CN110008863A - Efficient multi-scale sampling method based on high-resolution bridge area ship image - Google Patents
Efficient multi-scale sampling method based on high-resolution bridge area ship image Download PDFInfo
- Publication number
- CN110008863A CN110008863A CN201910220450.4A CN201910220450A CN110008863A CN 110008863 A CN110008863 A CN 110008863A CN 201910220450 A CN201910220450 A CN 201910220450A CN 110008863 A CN110008863 A CN 110008863A
- Authority
- CN
- China
- Prior art keywords
- ship
- sampling
- image
- length
- general image
- 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
Classifications
-
- 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
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of efficient multi-scale sampling methods based on high-resolution bridge area ship image, it solves the existing method of sampling and calculates the disadvantage that cost is excessively high, speed is excessively slow, it include: the overlapping dichotomy for proposing to be directed to high-definition picture, general image specific gravity, which is accounted for, according to minimum target to be measured determines the sampling number of plies, first layer is sampled as general image length-width direction respectively carrying out two points, subgraph length-width direction is respectively carried out two points on the basis of first layer by second layer sampling, is so recycled；To advanced optimize the overlapping dichotomy based on attention mechanism for proposing to be directed to high-resolution ship image, general image is decomposed using sparse low-order decomposition method, wherein sparse component is ship marking area figure, the candidate window generated in dichotomy will be overlapped to be ranked up according to the accounting comprising notable figure, generate the candidate window collection based on attention mechanism after determining interceptive value.The present invention is convenient, accurate, improves the efficiency of ship identification.
Description
Technical field
Present invention relates particularly to a kind of efficient multi-scale sampling methods based on high-resolution bridge area ship image.
Background technique
With the fast development of bridge construction industry and shipping industry, the contradiction between ship and bridge is more and more sharp.
Ship load is as a kind of very important important accidental load, once applying, will generate to the normal service of bridge huge
Threat.Therefore, whether for the bridge for being in the construction period or having built up, anti-ship, which hits early warning just, becomes it in weight
Weight.And the accurate identification of ship is exactly most important step in early warning.With the development of camera work, the navigation channel prison of bridge installation
Control camera often can all acquire thousands of pixels multiplied by the high-definition picture of thousands of pixels, and mainstream computer vision side now
Method is all based on low-resolution image, thus how by the algorithm of advanced ship identification apply on high-definition picture just at
The problem of being concerned.
Now with the development of computer vision, the ship that many scholars attempt to solve bridge area high-definition picture is known
Other problem, however these methods are the method based on sliding window mostly, the candidate window enormous amount that such methods generate,
A large amount of computing resource can be consumed, this is identified for the high-efficiency ship in high-definition picture increases difficulty.How for existing
The method that the problem of with the presence of research proposes an efficient multi-scale sampling provides one for the application of advanced computers vision technique
A feasible approach hits for the anti-ship in bridge area and provides the solution of an automation and intelligentification, is one and urgently studies
Problem.
Summary of the invention
The purpose of the present invention is to solve existing high-resolution bridge area ship image traditional multiscale transform sampling sides
Method calculates the disadvantages of cost is excessively high, speed is excessively slow, proposes a kind of efficient more rulers based on high-resolution bridge area ship image
The method of sampling is spent, the quick sampling of high-definition picture, the rapidly extracting of ship marking area and candidate window are realized
The application that intelligent selection is advanced ship identification technology on high-definition picture provides a feasible approach, is bridge
The automatic monitoring that the anti-ship of engineering is hit provides solution.
The technology used in the present invention is as follows: a kind of efficient based on high-resolution bridge area ship image multiple dimensioned is adopted
Quadrat method specifically comprises the following steps:
Step 1: accounting for general image specific gravity according to minimum target to be measured determines the sampling number of plies；
Step 2: sample to general image, first layer is sampled as respectively carrying out general image length-width direction two points,
Subgraph length-width direction is respectively carried out two points on the basis of first layer by two layers of sampling, is so recycled；
Step 3: general image is decomposed using sparse low-order decomposition method, wherein sparse component is that ship is aobvious
Write administrative division map；
Step 4: being ranked up the candidate window generated in dichotomy is overlapped according to the accounting comprising notable figure, determine
The candidate window collection based on attention mechanism is generated after interceptive value.
The present invention also has following technical characteristic:
1, it in step 1 as described above, determines that minimum target to be measured accounts for general image specific gravity according to sampling purpose, such as adopts
Sample purpose is target detection, and its input requirements is m × m, then minimum target is more than or equal to m × m, and sampling the number of plies is 1, small
In m × m more than or equal to m/10 × m/10, the sampling number of plies is 2, and less than or equal to m/10 × m/10, the sampling number of plies is 3.
2, in such as above-mentioned step two, general image (M × N) is respectively carried out two in length-width direction when first layer samples
Point, common property is raw 4 sub-regions (M/2 × N/2), and it is M/2 that length direction, which is overlapped step-length, and it is N/2 that width direction, which is overlapped step-length, so
Circulation.
3, in such as above-mentioned step three, the objective function of sparse low-order decomposition method is converted are as follows:
4, f in formulaD,fT,fBAnd fNRespectively whole ship image, ship component, background component and random noise component, λ
For parameter,It can guarantee ship component discomposing effect, carry out image two after the sparse component extraction of ship
Value processing, converts ship notable figure for ship component.
5, in such as above-mentioned step four, the selection of interceptive value both needs to consider effectively to cross noise filtering, considers to retain again
Small object, interceptive value 1%-5%.
The invention has the benefit that the present invention is convenient, accurate, the ship in the high-definition picture of bridge area is improved
The efficiency of identification.Entire sampling process is automatic processing, significantly reduces the artificial participation in detection process.The present invention
Bridge area high-definition picture vessel area is also able to satisfy to automatically extract and real time data processing demand.The present invention improves bridge
Automation, intelligence, accuracy and the robustness that vessel area in beam high-resolution region image is extracted, it is anti-for science of bridge building
The automatic monitoring that ship is hit provides feasible solution route.
Detailed description of the invention
Fig. 1 is flow diagram of the invention；
Fig. 2 is the sampled result figure of one embodiment of step 2 of the present invention；
Fig. 3 is the ship notable figure of one embodiment of step 3 of the present invention；
Fig. 4 is the efficient multi-scale sampling result of one embodiment of step 4 of the present invention.
Specific embodiment
Below according to Figure of description citing, the present invention will be further described:
Embodiment 1
A kind of efficient multi-scale sampling method based on high-resolution bridge area ship image, includes the following steps:
Step 1: according to minimum target to be measured account for general image specific gravity determine sampling the number of plies, according to sampling purpose determine to
It surveys minimum target and accounts for general image specific gravity, such as sampling purpose is target detection, and its input requirements is m × m, then minimum target is
More than or equal to m × m, sampling the number of plies is 1, is less than m × m more than or equal to m/10 × m/10, and the sampling number of plies is 2, is less than or equal to
M/10 × m/10's, the sampling number of plies is 3.
Step 2: sample to general image, first layer is sampled as respectively carrying out general image length-width direction two points,
Subgraph length-width direction is respectively carried out two points on the basis of first layer by two layers of sampling, is so recycled.First layer will be whole when sampling
Image (M × N) respectively carries out two points in length-width direction, and common property is raw 4 sub-regions (M/2 × N/2), and length direction overlapping step-length is
M/2, it is N/2 that width direction, which is overlapped step-length, is so recycled.
Step 3: general image is decomposed using sparse low-order decomposition method, wherein sparse component is that ship is aobvious
Administrative division map is write, low-rank sparse decomposition goal function can transform to:
F in formulaD,fT,fBAnd fNRespectively whole ship image, ship component, background component and random noise component, λ are
Parameter,It can guarantee ship component discomposing effect, carry out image two-value after the sparse component extraction of ship
Change processing, converts ship notable figure for ship component.
Step 4: being ranked up the candidate window generated in dichotomy is overlapped according to the accounting comprising notable figure, determine
The candidate window collection based on attention mechanism is generated after interceptive value.The selection of interceptive value, which both needs to consider effectively to filter, makes an uproar
Sound considers to retain Small object again, it is proposed that interceptive value 1%-5% is advisable.
Embodiment 2
As shown in Figure 1, a kind of efficient multi-scale sampling method based on high-resolution bridge area ship image, including such as
Lower step:
Step 1: according to minimum target to be measured account for general image specific gravity determine sampling the number of plies, according to sampling purpose determine to
It surveys minimum target and accounts for general image specific gravity, such as sampling purpose is target detection, and its input requirements is m × m, then minimum target is
More than or equal to m × m, sampling the number of plies is 1, is less than m × m more than or equal to m/10 × m/10, and the sampling number of plies is 2, is less than or equal to
M/10 × m/10's, the sampling number of plies is 3.
(1), in one embodiment, for the general image having a size of 6000 × 4000 pixels, input requirements are 300 × 300
Pixel；
(2), minimum target is calculated having a size of 40 × 40 pixels；
(3), determine that the sampling number of plies is 2.
Step 2: sample to general image, first layer is sampled as respectively carrying out general image length-width direction two points,
Subgraph length-width direction is respectively carried out two points on the basis of first layer by two layers of sampling, is so recycled.First layer will be whole when sampling
Image (M × N) respectively carries out two points in length-width direction, and common property is raw 4 sub-regions (M/2 × N/2), and length direction overlapping step-length is
M/2, it is N/2 that width direction, which is overlapped step-length, is so recycled.
(1), in one embodiment, M=6000, N=4000 carry out first time sampling, raw 4 sub-regions of common property
(3000 × 2000 pixel), length direction step-length are 3000, width direction 2000, raw 9 candidate windows of common property；
(2), it carries out second to sample, common property is raw 16 sub-regions (1500 × 1000 pixel), and length direction step-length is
1500, width direction 1000, raw 49 candidate windows of common property, as shown in Figure 2.
Step 3: general image is decomposed using sparse low-order decomposition method.
(1), λ takes 0.0129, carries out sparse low-rank decomposition to general image using optimization formula；
(2), binarization operation is carried out to the sparse component of ship and obtains the notable figure of one embodiment, as shown in Figure 3.
Step 4: being ranked up the candidate window generated in dichotomy is overlapped according to the accounting comprising notable figure, determine
The candidate window collection based on attention mechanism is generated after interceptive value.
(1), totally 58 candidate windows in one embodiment are arranged according to the accounting comprising notable figure in step 3
Sequence；
(2), the raw candidate window based on attention mechanism totally 25 of common property after interceptive value is 1% is determined, such as Fig. 4 institute
Show.
Claims (5)
1. a kind of efficient multi-scale sampling method based on high-resolution bridge area ship image, which is characterized in that including such as
Lower step:
Step 1: accounting for general image specific gravity according to minimum target to be measured determines the sampling number of plies；
Step 2: sampling to general image, first layer is sampled as general image length-width direction respectively carrying out two points, the second layer
Subgraph length-width direction is respectively carried out two points on the basis of first layer by sampling, is so recycled；
Step 3: general image is decomposed using sparse low-order decomposition method, wherein sparse component is the significant area of ship
Domain figure；
Step 4: being ranked up the candidate window generated in dichotomy is overlapped according to the accounting comprising notable figure, truncation is determined
The candidate window collection based on attention mechanism is generated after threshold value.
2. a kind of efficient multi-scale sampling side based on high-resolution bridge area ship image according to claim 1
Method, which is characterized in that in step 1, determine that minimum target to be measured accounts for general image specific gravity according to sampling purpose, such as sample purpose
For target detection, and its input requirements is m × m, then minimum target is more than or equal to m × m, and sampling the number of plies is 1, is less than m × m
More than or equal to m/10 × m/10, sampling the number of plies is 2, and less than or equal to m/10 × m/10, the sampling number of plies is 3.
3. a kind of efficient multi-scale sampling side based on high-resolution bridge area ship image according to claim 1
Method, which is characterized in that in step 2, general image M × N is respectively carried out two points in length-width direction when first layer samples, common property
Raw 4 sub-regions M/2 × N/2, it is M/2 that length direction, which is overlapped step-length, and it is N/2 that width direction, which is overlapped step-length, is so recycled.
4. a kind of efficient multi-scale sampling side based on high-resolution bridge area ship image according to claim 1
Method, which is characterized in that the objective function transformation of sparse low-order decomposition method in step 3 are as follows:
F in formulaD,fT,fBAnd fNRespectively whole ship image, ship component, background component and random noise component, λ are parameter,Image binaryzation processing is carried out after the sparse component extraction of ship, is converted ship for ship component and is shown
Write figure.
5. a kind of efficient multi-scale sampling side based on high-resolution bridge area ship image according to claim 1
Method, which is characterized in that interceptive value=1%-5% in step 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910220450.4A CN110008863A (en) | 2019-03-22 | 2019-03-22 | Efficient multi-scale sampling method based on high-resolution bridge area ship image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910220450.4A CN110008863A (en) | 2019-03-22 | 2019-03-22 | Efficient multi-scale sampling method based on high-resolution bridge area ship image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110008863A true CN110008863A (en) | 2019-07-12 |
Family
ID=67167865
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910220450.4A Pending CN110008863A (en) | 2019-03-22 | 2019-03-22 | Efficient multi-scale sampling method based on high-resolution bridge area ship image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110008863A (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544502A (en) * | 2013-10-29 | 2014-01-29 | 上海市城市建设设计研究总院 | High-resolution remote-sensing image ship extraction method based on SVM |
| CN104408708A (en) * | 2014-10-29 | 2015-03-11 | 兰州理工大学 | Global-local-low-rank-based image salient target detection method |
| CN107464255A (en) * | 2017-08-08 | 2017-12-12 | 大连海事大学 | A kind of ship target detection method based on information content Yu multiple dimensioned abnormality detection |
| CN108319949A (en) * | 2018-01-26 | 2018-07-24 | 中国电子科技集团公司第十五研究所 | Mostly towards Ship Target Detection and recognition methods in a kind of high-resolution remote sensing image |
-
2019
- 2019-03-22 CN CN201910220450.4A patent/CN110008863A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544502A (en) * | 2013-10-29 | 2014-01-29 | 上海市城市建设设计研究总院 | High-resolution remote-sensing image ship extraction method based on SVM |
| CN104408708A (en) * | 2014-10-29 | 2015-03-11 | 兰州理工大学 | Global-local-low-rank-based image salient target detection method |
| CN107464255A (en) * | 2017-08-08 | 2017-12-12 | 大连海事大学 | A kind of ship target detection method based on information content Yu multiple dimensioned abnormality detection |
| CN108319949A (en) * | 2018-01-26 | 2018-07-24 | 中国电子科技集团公司第十五研究所 | Mostly towards Ship Target Detection and recognition methods in a kind of high-resolution remote sensing image |
Non-Patent Citations (5)
| Title |
|---|
| M.O. ULFARSSON等: "Sparse and low rank decomposition using l0 penalty", 《2015 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING》 * |
| ZIHAN ZHOU等: "Stable Principal Component Pursuit", 《2010 IEEE INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY》 * |
| 李春雷等: "基于对偶范数低秩分解的织物疵点检测方法", 《棉纺织技术》 * |
| 汪闽 等: "高分辨率遥感影像上基于形状特征的船舶提取", 《武汉大学学报.信息科学版》 * |
| 陆玲 等: "《图像目标分割方法》", 30 November 2016, 哈尔滨工程大学出版社 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Autonomous damage segmentation and measurement of glazed tiles in historic buildings via deep learning | |
| Wang et al. | RENet: Rectangular convolution pyramid and edge enhancement network for salient object detection of pavement cracks | |
| Li et al. | Sewer pipe defect detection via deep learning with local and global feature fusion | |
| Wang et al. | Towards an automated condition assessment framework of underground sewer pipes based on closed-circuit television (CCTV) images | |
| CN110569738B (en) | Natural scene text detection method, equipment and medium based on densely connected network | |
| CN108596952B (en) | Rapid deep learning remote sensing image target detection method based on candidate region screening | |
| Sun et al. | Autonomous crack and bughole detection for concrete surface image based on deep learning | |
| Zhang et al. | An efficient lightweight convolutional neural network for industrial surface defect detection | |
| Xu et al. | Deep learning of rock microscopic images for intelligent lithology identification: Neural network comparison and selection | |
| Wu et al. | Applying deep convolutional neural network with 3D reality mesh model for water tank crack detection and evaluation | |
| Chen et al. | Quantify pixel-level detection of dam surface crack using deep learning | |
| CN112560895A (en) | Bridge crack detection method based on improved PSPNet network | |
| Zhang et al. | Investigation of pavement crack detection based on deep learning method using weakly supervised instance segmentation framework | |
| Altabey et al. | Research in image processing for pipeline crack detection applications | |
| Jia et al. | Intelligent identification of metal corrosion based on Corrosion-YOLOv5s | |
| Fan et al. | Pavement Defect Detection with Deep Learning: A Comprehensive Survey | |
| Chen et al. | A hierarchical DCNN-based approach for classifying imbalanced water inflow in rock tunnel faces | |
| Zhao et al. | A method of crack detection based on convolutional neural networks | |
| Nguyen et al. | Automatic measurement of concrete crack width in 2D multiple-phase images for building safety evaluation | |
| CN113139511A (en) | Water and soil conservation production and construction project disturbance pattern spot extraction method, medium and equipment | |
| Zhu | Detection and recognition of concrete cracks on building surface based on machine vision | |
| Liu et al. | Research on surface defect detection based on semantic segmentation | |
| Yuan et al. | Automated pixel-level crack detection and quantification using deep convolutional neural networks for structural condition assessment | |
| CN112084941A (en) | Target detection and identification method based on remote sensing image | |
| CN110008863A (en) | Efficient multi-scale sampling method based on high-resolution bridge area ship image |
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: 20190712 |
|
| RJ01 | Rejection of invention patent application after publication |