CN112241950A - Detection method of tower crane crack image - Google Patents
Detection method of tower crane crack image Download PDFInfo
- Publication number
- CN112241950A CN112241950A CN202011117717.6A CN202011117717A CN112241950A CN 112241950 A CN112241950 A CN 112241950A CN 202011117717 A CN202011117717 A CN 202011117717A CN 112241950 A CN112241950 A CN 112241950A
- Authority
- CN
- China
- Prior art keywords
- tower crane
- crack
- crack image
- image
- data set
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to a detection method of a tower crane crack image, which comprises the following steps: step S1, acquiring a crack image of the tower crane, preprocessing the crack image, and dividing the crack image into a training set and a testing set; step S2, constructing an improved Faster R-CNN model; step S3, learning and training the improved Faster R-CNN model by adopting a transfer learning method based on the training set to obtain a trained Faster R-CNN model; step S4, fine-tuning the trained Faster R-CNN model to obtain a final detection model; and step S5, inputting the image to be detected into the detection model to obtain the position of the crack. The method can accurately identify the cracks of various scales, and the regression positioning is more accurate.
Description
Technical Field
The invention relates to the field of image detection, in particular to a method for detecting a crack image of a tower crane.
Background
In the construction of a construction site, the tower crane is overloaded for a long time, frequently operated, high in position and large in impact force, and is influenced by the natural environment for a long time, and the tower body structure is easy to crack, so that the bearing capacity is insufficient, and certain construction risk is caused. According to statistics, 655 hoisting machinery responsibility accidents of buildings in 2000-2012 occur in China, wherein 472 tower crane safety accidents occur; in 2012-2014, tower crane accidents occurred at 162, wherein the collapse and breakage accidents accounted for 123, which accounted for 75.93% of the total accidents, and mainly resulted from insufficient bearing capacity and unbalance due to cracking and cracks of the metal structure of the tower crane. The steel structure of tower crane adopts manual gas shielded welding basically when the welding, receives condition and personnel restriction, produces the welding seam defect very easily, and tower crane mainly is in the complex environment in the building site simultaneously, and the dead weight is up to hundreds of tons, and operating time has been of a specified duration and has appeared the problem of split and decrease easily, because tower crane's structure is many, the body of the tower is high, the crack of appearance is difficult to in time discover, and these crack defects can weaken tower crane's bearing capacity, bury the potential safety hazard for the building site construction. In recent years, since safety accidents of tower cranes have been more and more generated, strict management of tower cranes is necessary, and it is very important to ensure reliability and safety of tower cranes. The existing safety management of the tower crane is mainly based on manual inspection, but the manual detection of cracks of the tower crane has subjective influence, some small-scale cracks and some cracks with higher structure positions are easy to omit, a large amount of manpower, financial resources and energy are consumed, the efficiency and the accuracy are difficult to ensure, and more technical supports are needed.
Disclosure of Invention
In view of this, the present invention provides a method for detecting a crack image of a tower crane, which has a better detection effect and a higher detection accuracy, and is applied to detection of a crack image of a tower crane.
In order to achieve the purpose, the invention adopts the following technical scheme:
a detection method of a tower crane crack image comprises the following steps:
step S1, acquiring a crack image of the tower crane, preprocessing the crack image, and dividing the crack image into a training set and a testing set;
step S2, constructing an improved Faster R-CNN model;
step S3, learning and training the improved FasterR-CNN model by adopting a transfer learning method based on the training set to obtain a trained FasterR-CNN model;
step S4, fine-tuning the trained Faster R-CNN model to obtain a final detection model;
and step S5, inputting the image to be detected into the detection model to obtain the position of the crack.
Further, the step S1 is specifically:
step S11, acquiring tower crane crack images with different illumination effects, angles, shapes and sizes to obtain a tower crane crack data set;
step S12, translating, zooming and rotating the tower crane crack image, expanding the tower crane crack data set, and adding the railway crack image, the highway crack image and the bridge crack image to obtain an expanded data set;
and step S13, labeling the crack targets by using a LabelImg tool on the expanded data set, converting the crack targets into a VOC 2007 data set format, and dividing the crack targets into a training set and a data set according to a preset proportion.
Further, the improved Faster R-CNN model adopts a multi-scale characteristic pyramid structure to replace an original RPN structure, and the number and the size of anchor frames are improved; and the RoI Pooling layer is modified to introduce the RoI Align method.
Furthermore, the multi-scale feature pyramid structure is based on ResNet-50 and is realized by combining with an FPN algorithm.
Further, the multi-scale feature pyramid structure comprises three parts, namely a channel from bottom to top, a channel from top to bottom and a transverse connection part, and the three parts are as follows:
for the input picture, the characteristic diagram extraction mode adopts from bottom to top;
2 times of upsampling is realized through a top-down method, and then the upsampling and the feature map are subjected to feature fusion;
the cross-concatenation is performed by performing a convolution operation of 1x1 on the feature map generated from bottom-up.
Further, the number and size of the improved anchor frame are specifically as follows: at the original three different scales and three ratios 1: 1,1: 2,2: 1, adding anchor points with the scales of 32x32 and 64x64, and modifying the anchor to five different scales of 32x32, 64x64, 128x128, 256x256 and 512x512, wherein the original scale is adopted as 1: 1,1: 2,2: 1, combining into 15 anchors with different sizes and different length-width ratios.
Further, the improved RoI Pooling layer specifically comprises: a RoI Align method provided in a Mask R-CNN algorithm is adopted to improve a RoI Pooling layer, a bilinear interpolation method is adopted, and floating point numbers are reserved.
A tower crane crack image detection system comprising a memory having stored thereon a computer program enabling a processor to carry out the method steps according to any one of claims 1-7 when the computer program is run.
Compared with the prior art, the invention has the following beneficial effects:
1. the method can accurately identify the cracks of various scales, and the regression positioning is more accurate.
2. The method has better detection effect and higher detection precision, and realizes the detection of the crack image of the tower crane.
Drawings
FIG. 1 is a flow chart of the operation of an embodiment of the present invention;
FIG. 2 illustrates an RPN structure according to an embodiment of the present invention;
FIG. 3 illustrates a multi-scale feature pyramid structure in accordance with an embodiment of the present invention;
FIG. 4 is a flow diagram of a transfer learning application in accordance with an embodiment of the present invention;
FIG. 5 is a diagram illustrating the detection effect according to an embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
Referring to fig. 1, the present invention provides a method for detecting a crack image of a tower crane, including the following steps:
step S1, acquiring a crack image of the tower crane, preprocessing the crack image, and dividing the crack image into a training set and a testing set;
step S2, constructing an improved Faster R-CNN model;
step S3, learning and training the improved Faster R-CNN model by adopting a transfer learning method based on the training set to obtain a trained Faster R-CNN model;
step S4, fine-tuning the trained Faster R-CNN model to obtain a final detection model;
and step S5, inputting the image to be detected into the detection model to obtain the position of the crack.
In this embodiment, the step S1 specifically includes:
step S11, acquiring tower crane crack images with different illumination effects, angles, shapes and sizes to obtain a tower crane crack data set;
step S12, translating, zooming and rotating the tower crane crack image, expanding the tower crane crack data set, and adding the railway crack image, the highway crack image and the bridge crack image to obtain an expanded data set;
and step S13, labeling the crack targets by using a LabelImg tool on the expanded data set, converting the crack targets into a VOC 2007 data set format, and dividing the crack targets into a training set and a data set according to a preset proportion.
In the implementation, the CNN feature extraction network of the improved Faster R-CNN model is based on ResNet-50 and combined with an FPN algorithm to realize a multi-scale feature pyramid structure to replace an original RPN structure. ResNet-50 can be used for solving the problem of difficulty in deep network training, a bottleeck structure is introduced, the number of channels is reduced by utilizing convolution operation of 1x1, so that the calculated amount can be reduced, layer jump connection is realized through identity mapping, the convergence speed is accelerated, and the difficulty in optimizing a network model is reduced.
The RPN structure is a full convolution network, and is mapped into d-dimensional feature vectors by performing sliding window processing on a feature map output by a convolution layer, and then the d-dimensional feature vectors are input into a classification layer and a regression layer to judge the category and the position of the next step. And simultaneously predicting k candidate regions at the position of each sliding window, wherein the regression layer has 4k outputs and represents 4 coordinates of k candidate frames, and the classification layer has 2k score outputs and represents the estimated probability of whether each candidate frame is a target. In order to enable the algorithm to be better suitable for targets with different sizes and shapes, a plurality of anchor boxes are arranged at each position on the feature map, so that the anchor boxes are used for predicting candidate regions with different length-width ratios and different scales of the input image. The overall loss (loss) function when training the RPN is:
wherein i represents the ith anchor, piIs the probability that the anchor contains the target; when anchor is a positive sample, pi *If it is negative, then p is 1i *＝0；tiIs a coordinate vector used to predict the candidate box, ti＝(tx,ty,tw,th)，ti *Then corresponds to the real target box, ti *＝(tx *,ty *,tw *,th *)，tiAnd ti *Is defined as:
wherein x, y, w, h are the center coordinates and width and height of the RPN prediction box, xa,ya,wa,haCorresponding to the anchor box, x*,y*,w*,h*Corresponding to the real frame; n is a radical ofclsIs the size of the batch, NregLambda is the balance weight, given by N, for the total number of anchorsclsAnd NregAnd λ is normalized. L isclsIs a classification loss function defined as:
Lcls(pi,pi *)＝-log[pi *pi+(1-pi *)(1-pi)]
Lregis a regression loss function defined as:
wherein, SmoothL1For the robust loss function, defined as:
referring to fig. 3, in the present embodiment, preferably, the multi-scale feature pyramid algorithm is integrally divided into three parts: the channels from bottom to top, the channels from top to bottom and the channels are transversely connected. Firstly, inputting a picture, wherein the characteristic diagram extraction mode adopts a bottom-up mode. Secondly, 2 times of upsampling is realized through a top-down method, and then feature fusion is realized with a feature map. Finally, the horizontal connection is carried out, the convolution operation of 1x1 is carried out on the feature map generated from bottom to top, the aim is to have the same channel number with the result of the up sampling, and then the convolution operation of 3x3 is carried out on each fusion result, so that the aliasing effect caused by the up sampling is eliminated. Through the operation, the fused feature layers are (P2, P3, P4 and P5), and more superficial feature target position detail information is fused. And then generating a candidate frame through the RPN network, and further judging the category and the position of the target. Combining with FPN algorithm, a plurality of characteristic graphs can be obtained, and different characteristic layers are selected by the RoI according to the characteristic graphs with different scales, wherein the specific selection mode is as follows:
wherein k is05, representing the output of layer P5; w and h respectively represent the width and height of the RoI region; 224 is a standard input for image data sets.
Secondly, improving the anchor frame:
under original parameters, anchor points with the scales of 32x32 and 64x64 are added, and the anchors are improved to be five different scales (32x32, 64x64, 128x128, 256x256 and 512x512) respectively, the original proportions (1: 1, 1: 2 and 2: 1) are adopted in proportion, and the anchor points are combined into 15 anchors with different scales and different length-width ratios.
(iii) modification of the RoI Pooling layer:
a RoI Align method provided in a Mask R-CNN algorithm is adopted to improve a RoI Pooling layer, a bilinear interpolation method is adopted to reserve floating point numbers, and precision loss caused by quantization is avoided. Setting the size of an input image as 900x900, assuming that the size of a target size in a crack image of the tower crane to be detected is 330x330, performing maximum pooling for 4 times when passing through a feature extraction network, reducing an area suggestion frame to 1/16 of an original image, and changing the size to 20.63x 20.63; after the first quantization, the RoI Pooling is rounded, the area suggestion box size is changed into 20x20, the RoI Align retains floating point numbers, and the size is still 20.63x 20.63; then Pooling the candidate regions, fixing the size to 7x7, performing second quantization, taking 20/7 as 2.86 and 20/7 as 2.86 by RoI Pooling, wherein the size of the feature subgraph is 2x2 after rounding, taking 20.63/7 as 2.95 and 20.63/7 as 2.95 by RoI Align, and keeping the size of the feature subgraph to 2.95x2.95 after floating point numbers are reserved; and finally, taking the maximum value of each feature sub-graph and outputting the maximum value to form 49 pixel values, and forming a 7x7 feature graph. The RoI Pooling rounds the whole, the area of the map on the feature map suggests that the box size is changed from 20x20 to 14x14, which has large pixel deviation and is not beneficial to the regression positioning operation. Therefore, the accuracy of regression positioning can be better improved by using the RoI Align method, so that a more accurate target region is obtained.
Referring to fig. 4, the migration learning is specifically as follows, using ResNet-50 pre-trained by ImageNet to perform model fine tuning, so as to obtain a model suitable for tower crane crack detection, so that a good recognition result can be achieved with only a small amount of data sets of tower crane cracks. And transferring the parameters of the initial model obtained by pre-training to a target data set and then re-learning. In order to prevent the parameters from being excessively distorted in the subsequent transfer learning process, the SGD algorithm is used for optimizing the parameters, the initial learning rate of the transfer learning model is set to be 0.001, the batch size is set to be 1, the momentum factor is set to be 0.9, the weight attenuation factor is set to be 0.0005, the iteration number is set to be 20000, and then the model is finely adjusted to obtain the final model.
A tower crane crack image detection system comprising a memory having stored thereon a computer program enabling a processor to carry out the method steps according to any one of claims 1-7 when the computer program is run.
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.
Claims (8)
1. A detection method for a tower crane crack image is characterized by comprising the following steps:
step S1, acquiring a crack image of the tower crane, preprocessing the crack image, and dividing the crack image into a training set and a testing set;
step S2, constructing an improved Faster R-CNN model;
step S3, learning and training the improved Faster R-CNN model by adopting a transfer learning method based on the training set to obtain a trained Faster R-CNN model;
step S4, fine-tuning the trained Faster R-CNN model to obtain a final detection model;
and step S5, inputting the image to be detected into the detection model to obtain the position of the crack.
2. The method for detecting the tower crane crack image according to claim 1, wherein the step S1 specifically comprises:
step S11, acquiring tower crane crack images with different illumination effects, angles, shapes and sizes to obtain a tower crane crack data set;
step S12, translating, zooming and rotating the tower crane crack image, expanding the tower crane crack data set, and adding the railway crack image, the highway crack image and the bridge crack image to obtain an expanded data set;
and step S13, labeling the crack targets by using a LabelImg tool on the expanded data set, converting the crack targets into a VOC 2007 data set format, and dividing the crack targets into a training set and a data set according to a preset proportion.
3. The method for detecting the crack image of the tower crane according to claim 1, wherein the improved Faster R-CNN model adopts a multi-scale feature pyramid structure to replace an original RPN structure, and the number and the size of anchor frames are improved; and the RoI Pooling layer is modified to introduce the RoI Align method.
4. The method for detecting the tower crane crack image according to claim 3, wherein the multi-scale feature pyramid structure is a multi-scale feature pyramid structure realized based on ResNet-50 and combined with an FPN algorithm.
5. The method for detecting the tower crane crack image according to claim 4, wherein the multi-scale feature pyramid structure comprises three parts, namely a bottom-up channel, a top-down channel and a transverse connection part, and the method comprises the following specific steps:
for the input picture, the characteristic diagram extraction mode adopts from bottom to top;
2 times of upsampling is realized through a top-down method, and then the upsampling and the feature map are subjected to feature fusion;
the cross-concatenation is performed by performing a convolution operation of 1x1 on the feature map generated from bottom-up.
6. The method for detecting the tower crane crack image according to claim 3, wherein the number and the size of the improved anchor frames are specifically as follows: at the original three different scales and three ratios 1: 1,1: 2,2: 1, adding anchor points with the scales of 32x32 and 64x64, and modifying the anchor to five different scales of 32x32, 64x64, 128x128, 256x256 and 512x512, wherein the original scale is adopted as 1: 1,1: 2,2: 1, combining into 15 anchors with different sizes and different length-width ratios.
7. The method for detecting the tower crane crack image according to claim 3, wherein the improved RoI Pooling layer is specifically as follows: a RoI Align method provided in a Mask R-CNN algorithm is adopted to improve a RoI Pooling layer, a bilinear interpolation method is adopted, and floating point numbers are reserved.
8. A tower crane crack image detection system comprising a memory having a computer program stored thereon, and a processor capable of performing the method steps of any one of claims 1-7 when the computer program is executed by the processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011117717.6A CN112241950A (en) | 2020-10-19 | 2020-10-19 | Detection method of tower crane crack image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011117717.6A CN112241950A (en) | 2020-10-19 | 2020-10-19 | Detection method of tower crane crack image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112241950A true CN112241950A (en) | 2021-01-19 |
Family
ID=74169064
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011117717.6A Pending CN112241950A (en) | 2020-10-19 | 2020-10-19 | Detection method of tower crane crack image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112241950A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112818878A (en) * | 2021-02-05 | 2021-05-18 | 福州大学 | Tower crane crack identification and qualitative method |
| CN113111722A (en) * | 2021-03-17 | 2021-07-13 | 天津理工大学 | Automatic driving target identification method based on improved Mask R-CNN |
| CN113255626A (en) * | 2021-07-14 | 2021-08-13 | 杭州大杰智能传动科技有限公司 | Intelligent tower crane structure state detection method and device based on scanned image analysis |
| CN113537045A (en) * | 2021-07-14 | 2021-10-22 | 宁夏大学 | Rock picture detection method based on improved FasterR-CNN |
| CN114742822A (en) * | 2022-05-20 | 2022-07-12 | 青岛农业大学 | Construction method and application of strawberry identification and counting model |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103453892A (en) * | 2013-09-24 | 2013-12-18 | 广东工业大学 | Bridge crack real-time monitoring and quick pre-warning device |
| US20190154442A1 (en) * | 2016-01-15 | 2019-05-23 | Fugro Roadware Inc. | High speed stereoscopic pavement surface scanning system and method |
| KR102008973B1 (en) * | 2019-01-25 | 2019-08-08 | (주)나스텍이앤씨 | Apparatus and Method for Detection defect of sewer pipe based on Deep Learning |
| CN110211097A (en) * | 2019-05-14 | 2019-09-06 | 河海大学 | A kind of crack image detecting method based on the migration of Faster R-CNN parameter |
| CN110321815A (en) * | 2019-06-18 | 2019-10-11 | 中国计量大学 | A kind of crack on road recognition methods based on deep learning |
| CN111582339A (en) * | 2020-04-28 | 2020-08-25 | 江西理工大学 | Vehicle detection and identification method based on deep learning |
-
2020
- 2020-10-19 CN CN202011117717.6A patent/CN112241950A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103453892A (en) * | 2013-09-24 | 2013-12-18 | 广东工业大学 | Bridge crack real-time monitoring and quick pre-warning device |
| US20190154442A1 (en) * | 2016-01-15 | 2019-05-23 | Fugro Roadware Inc. | High speed stereoscopic pavement surface scanning system and method |
| KR102008973B1 (en) * | 2019-01-25 | 2019-08-08 | (주)나스텍이앤씨 | Apparatus and Method for Detection defect of sewer pipe based on Deep Learning |
| CN110211097A (en) * | 2019-05-14 | 2019-09-06 | 河海大学 | A kind of crack image detecting method based on the migration of Faster R-CNN parameter |
| CN110321815A (en) * | 2019-06-18 | 2019-10-11 | 中国计量大学 | A kind of crack on road recognition methods based on deep learning |
| CN111582339A (en) * | 2020-04-28 | 2020-08-25 | 江西理工大学 | Vehicle detection and identification method based on deep learning |
Non-Patent Citations (3)
| Title |
|---|
| LIN T Y ETAL.: "Feature Pyramid Networks for Object Detection", 《2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
| 张神德: "塔式起重机安全管理中裂缝检测方法", 《福建电脑》 * |
| 李太文: "基于Faster R—CNN的道路裂缝识别", 《电子技术应用》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112818878A (en) * | 2021-02-05 | 2021-05-18 | 福州大学 | Tower crane crack identification and qualitative method |
| CN113111722A (en) * | 2021-03-17 | 2021-07-13 | 天津理工大学 | Automatic driving target identification method based on improved Mask R-CNN |
| CN113255626A (en) * | 2021-07-14 | 2021-08-13 | 杭州大杰智能传动科技有限公司 | Intelligent tower crane structure state detection method and device based on scanned image analysis |
| CN113537045A (en) * | 2021-07-14 | 2021-10-22 | 宁夏大学 | Rock picture detection method based on improved FasterR-CNN |
| CN113537045B (en) * | 2021-07-14 | 2022-11-25 | 宁夏大学 | Rock picture detection method based on improved FasterR-CNN |
| CN114742822A (en) * | 2022-05-20 | 2022-07-12 | 青岛农业大学 | Construction method and application of strawberry identification and counting model |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112241950A (en) | Detection method of tower crane crack image | |
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| CN110503112B (en) | Small target detection and identification method for enhancing feature learning | |
| CN112767357A (en) | Yolov 4-based concrete structure disease detection method | |
| CN109446992A (en) | Remote sensing image building extracting method and system, storage medium, electronic equipment based on deep learning | |
| CN111179217A (en) | Attention mechanism-based remote sensing image multi-scale target detection method | |
| CN112700444B (en) | Bridge bolt detection method based on self-attention and central point regression model | |
| CN111127416A (en) | Computer vision-based automatic detection method for surface defects of concrete structure | |
| CN109858367B (en) | Visual automatic detection method and system for worker through supporting unsafe behaviors | |
| CN108009509A (en) | Vehicle target detection method | |
| Li et al. | Automatic bridge crack identification from concrete surface using ResNeXt with postprocessing | |
| CN113822880A (en) | Crack identification method based on deep learning | |
| CN113378686A (en) | Two-stage remote sensing target detection method based on target center point estimation | |
| CN115409069A (en) | Village and town building identification method, classification method, device, electronic equipment and medium | |
| CN113033315A (en) | Rare earth mining high-resolution image identification and positioning method | |
| CN113468968A (en) | Remote sensing image rotating target detection method based on non-anchor frame | |
| CN114494870A (en) | Double-time-phase remote sensing image change detection method, model construction method and device | |
| CN112101138A (en) | Bridge inhaul cable surface defect real-time identification system and method based on deep learning | |
| CN113902792A (en) | Building height detection method and system based on improved RetinaNet network and electronic equipment | |
| CN113052217A (en) | Prediction result identification and model training method and device thereof, and computer storage medium | |
| CN112733756A (en) | Remote sensing image semantic segmentation method based on W divergence countermeasure network | |
| CN112215301A (en) | Image straight line detection method based on convolutional neural network | |
| CN114494893B (en) | Remote sensing image feature extraction method based on semantic reuse context feature pyramid | |
| CN110889418A (en) | Gas contour identification method | |
| CN113673478B (en) | Port large-scale equipment detection and identification method based on deep learning panoramic stitching |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210119 |
|
| RJ01 | Rejection of invention patent application after publication |