CN102867196B - Based on the complicated sea remote sensing image Ship Detection of Gist feature learning - Google Patents
Based on the complicated sea remote sensing image Ship Detection of Gist feature learning Download PDFInfo
- Publication number
- CN102867196B CN102867196B CN201210339791.1A CN201210339791A CN102867196B CN 102867196 B CN102867196 B CN 102867196B CN 201210339791 A CN201210339791 A CN 201210339791A CN 102867196 B CN102867196 B CN 102867196B
- Authority
- CN
- China
- Prior art keywords
- image
- feature
- brightness
- remote sensing
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
Based on a complicated sea remote sensing image Ship Detection for Gist feature learning, comprise the following steps: step 1, gather the complicated sea remote sensing image data of different phase, different sensors, different scale; Step 2, carries out partitioning pretreatment to complicated sea remote sensing image, obtains sample image section and detects image slice; Step 3, extracts notable feature and the Gist feature of sample image section and detection image slice; Step 4, trains sample image section according to step 3 gained notable feature and Gist feature, obtains training pattern; Step 5, according to step 4 gained training pattern, judges whether to detect image slice containing naval vessel by SVM classifier; Step 6, finds the single naval vessel detecting image slice based on the itti visual attention model improved.The present invention reduces its false alarm rate while ensure that not undetected naval vessel as far as possible, and effectively can process the complicated sea remote sensing image containing the disturbed condition such as sea clutter, cloud and mist, computation complexity is low, with strong points.
Description
Technical field
The present invention relates to Remote Sensing Image Processing Technology field, especially relate to a kind of complicated sea remote sensing image Ship Detection based on Gist biological vision feature learning.
Background technology
Ship Target Detection is the task with traditional of each coastal strip country of the world, find with relief on naval vessel, fisherman monitor, illegal immigrant, defendance territory, anti-drug, naval vessel illegal dumping greasy dirt Monitoring and Management etc. in have a wide range of applications.Along with the development of remotely sensed image technology, utilizing remote sensing images to carry out Ship Target Detection becomes possibility, and its research object comprises the detection to naval vessel itself and ship Wake.Under complicated marine environmental conditions, satellite remote-sensing image presents rambling fish scale light, large area retroreflective regions, irregular movement enrich texture wave etc., middle-size and small-size Ship Target may be hidden in complicated background clutter, thus affects ship seakeeping effect.On Ship Target image, sea situation (stormy waves), meteorological (cloud and mist), water colour etc. cause the interference such as remote sensing image Zhong Hai land characteristic instability, wave of the sea, floating on water thing, ship Wake very many, and naval vessel detection difficulty is large.
Existing Ship Target Detection mainly contains following algorithm: based on the algorithm of target detection etc. in the algorithm of global threshold segmentation, algorithm based on Local threshold segmentation, best entropy automatic threshold method, the algorithm based on distributed model, the algorithm based on fractal model, feature based territory, but their ubiquities detect the too high problem of false alarm rate, and universality is poor.Wherein, global threshold algorithm cannot regulate threshold value automatically according to the change of regional area in image, therefore testing result is subject to localized variation and introduces a large amount of false-alarm and undetected, the image wider to fabric width, due to the reason of imaging mechanism, Sea background has strong grey scale change, also can cause occurring false-alarm and undetected.These class methods only make use of the gray-scale statistical characteristic of target in testing process, do not consider the spatial structural form of target, and histogram shape and contacting of picture material also have uncertainty; Local threshold algorithm spot make an uproar more, sea stormy waves larger time, easily cause a large amount of false-alarms.This algorithm is owing to needing repeatedly the statistical parameter of background area in calculation window simultaneously, and operand is very big, and the speed of processing procedure is comparatively slow, can not meet the needs of real-time or near process in real time in practical application; Choosing of best entropy automatic threshold method threshold value only make use of gray-scale statistical characteristic, does not consider the spatial structural form of target, and histogrammic distribution and contacting of picture material also have uncertainty.In traditional KSW algorithm, criterion function is simply defined as target gray entropy and background gray level entropy sum, and background gray level entropy and target gray entropy occupy equal ratio in criterion function.This have ignored background and target proportion different on image, also have ignored background and the difference of target in tonal range.When image comprises stronger sea clutter, the testing result that often can not get; First algorithm based on distributed model needs hypothesis background being done to clutter distribution, and this just needs certain priori, but in fact generally background clutter does not strictly obey certain distribution yet.Secondly, such algorithm needs to add up pixel each in image, and therefore calculated amount is comparatively large, and increases along with the increase of sliding window size; Algorithm based on fractal model thinks that the fractal dimension of natural scene and Ship Target has certain difference, detects according to difference.But affect by background complexity, random noise, image quality etc. in real image, single yardstick or constant fractal dimension are difficult to distinguish natural scene and man-made target; The algorithm of target detection in feature based territory, when the intensity profile of background is more complicated, greatly affected by noise, now can increase the impact of noise on characteristic pattern, produces segmentation by mistake.In addition, in the process of Feature Conversion, also some can be had to affect on the profile of target itself, utilize the method segmentation object, shape information will be lost to some extent.
Can find out that various algorithm still will be subject to the restriction of all many condition, the interference of such as image background, target are subject to the impact of weather, illumination variation, particularly for the remotely-sensed data of middle low resolution, Ship Target is rendered as Small object on image, occurs that the probability of false dismissal, false-alarm is higher in observation process.Simultaneously, in the ordinary course of things, visible images can be subject to cloud layer and the interference such as greasy dirt, wave, set up background model more difficult, in image, object and background otherness is inconsistent, and in image, naval vessel intensity profile is uneven, also not obvious with extra large background contrast, therefore also not easily split, especially for black polarity naval vessel, all the more so.
The formula of discrete square conversion (DMT) of prior art
In formula, i, j detect ranks value in image slice, and g (i-r, j-s) represents the original pixel value detecting the capable jth of the i-th-r-s row in image slice; K is default window size parameter, such as k=1, and window is 3 × 3; R, s are the loop variables in window, and p, q are indexes, and during concrete enforcement, those skilled in the art can set k, r, s value voluntarily as required.Embodiment gets (p=0, q=1) (p=1, q=0) respectively, and (p=1, q=1) be i.e. three DMT textural characteristics, DMT
1.0dMT
0.1dMT
1.1.
Step 6.2, remarkable figure generates, comprise step 6.1 is obtained brightness, color, direction, texture 4 features by central peripheral difference operation generate each feature subgraph, then brightness is significantly schemed, color is significantly schemed, direction is significantly schemed, texture is significantly schemed to merge generation by local nonlinearity fusion method, finally merge the overall significantly figure of generation by local nonlinearity fusion method, in fusion generation, brightness is significantly schemed, color is significantly schemed, direction is significantly schemed, texture carries out square operation to each pixel value of individual features subgraph before significantly scheming.Due to square operation, the scope of pixel value in the remarkable figure that stretched, makes marking area layering more obvious, highlights Ship Target.
Summary of the invention
For the problems referred to above, the present invention proposes a kind of complicated sea remote sensing image Ship Detection based on Gist feature learning.
Technical scheme of the present invention is a kind of complicated sea remote sensing image Ship Detection based on Gist feature learning, comprises the following steps:
Step 1, gathers the data of complicated sea remote sensing image;
Step 2, carries out partitioning pretreatment to complicated sea remote sensing image, obtains sample image section and detects image slice;
Step 3, extracts notable feature and the Gist feature of sample image section and detection image slice;
Step 4, trains sample image section according to step 3 gained notable feature and Gist feature, obtains training pattern;
By SVM classifier, step 5, according to step 4 gained training pattern, judges that whether arbitrary detection image slice is containing naval vessel, is enter step 5, otherwise terminates the detection to this detection image slice;
Step 6, finds the single naval vessel detecting image slice based on itti visual attention model, obtain ship target.
And carry out notable feature extraction to arbitrary sample image section and detection image slice in step 3, implementation is as follows,
Image slice decomposed on color, brightness, these 3 feature passages of direction, color, brightness passage represent with gaussian pyramid respectively, generate the characteristic pattern on color characteristic passage and the characteristic pattern on brightness passage by the operation of central peripheral difference; Direction character passage Gabor filtering obtains 0 °, 45 °, 90 °, the direction pyramid in 135 ° of 4 directions, is generated the characteristic pattern on each direction by the operation of central peripheral difference;
Merged by local nonlinearity and the feature subgraph of each characteristic pattern is fused to color is significantly schemed, brightness is significantly schemed and direction is significantly schemed, finally represent the notable feature of image slice by the average of each remarkable figure, standard deviation, Local modulus maxima and Local modulus maxima spacing.
And carry out described Gist feature extraction to arbitrary sample image section and detection image slice in step 2, implementation is as follows,
Each image slice is decomposed on color, brightness, 3, direction feature passage, color, brightness passage use this pyramid representation of 9 floor heights respectively, generate the characteristic pattern on color characteristic passage and the characteristic pattern on brightness passage by the operation of central peripheral difference; Direction character passage Gabor filtering obtains 0 °, 45 °, 90 °, and the direction pyramid in 135 ° of 4 directions obtains the characteristic pattern on the passage of each direction;
The sub-grid that feature subgraph is divided into 4 × 4 sizes is often opened to each characteristic pattern, average after asking the average of 16 sub-grids, 4, upper left corner sub-grid, 4, upper right corner sub-grid, 4, lower left corner sub-grid, 4, upper left corner sub-grid, 4, upper left corner sub-grid to merge respectively respectively and the average of whole feature subgraph, finally represent Gist proper vector with acquired results.
And, find the single naval vessel detecting image slice in step 6 based on the itti visual attention model improved, comprise the following steps,
Step 6.1, carries out low-level visual features extraction to detection image slice, comprises brightness, color, direction and textural characteristics;
Step 6.2, remarkable figure generates, comprise step 6.1 is obtained brightness, color, direction, texture 4 features by central peripheral difference operation generate each feature subgraph, then brightness is significantly schemed, color is significantly schemed, direction is significantly schemed, texture is significantly schemed to merge generation by local nonlinearity fusion method, finally merges by local nonlinearity fusion method and generates overall remarkable figure; In fusion generation, brightness is significantly schemed, color is significantly schemed, direction is significantly schemed, texture carries out square operation to each pixel value of individual features subgraph before significantly scheming;
Step 6.3, Ship Target Detection, comprises the single naval vessel remarkable figure obtained in step 6.2 being detected to image slice by inhibition of return technology for detection, obtains ship target.
And, the remote sensing image data deriving from different phase, different sensors, different scale is selected when gathering the data of complicated sea remote sensing image, after carrying out partitioning pretreatment to complicated sea remote sensing image, the positive sample image of mark containing naval vessel and the negative sample image not containing naval vessel are cut into slices as sample image.
First the method that the present invention proposes carries out piecemeal to image, extract notable feature and the Gist feature of each sub-image block, proper vector according to extracting is trained sample data, then SVM classifier is utilized to judge whether containing naval vessel in sub-image block, finally based on the single naval vessel in the itti visual attention model searching sub-image block improved.The present invention reduces its false alarm rate while ensure that not undetected naval vessel as far as possible, and effectively can process the complicated sea remote sensing image containing the interference such as sea clutter, cloud and mist, computation complexity is low, with strong points; Be applicable to the complicated sea remote sensing image data of different phase, different resolution, there is good universality; Fast processing can be carried out to the mass remote sensing image data of large format and detect naval vessel.
Accompanying drawing explanation
Fig. 1 is process flow diagram of the present invention.
Embodiment
Technical solution of the present invention is described in detail below in conjunction with drawings and Examples.
The complicated sea remote sensing image Ship Detection that the present invention is based on Gist biological vision feature learning is the notable feature and the Gist feature that utilize naval vessel, the sub-image block SVM classifier (support vector machine classifier) of jumbo image is trained, obtain a forecast model, support vector in this model represents the characteristic feature on naval vessel, then the sub-image block having doubtful naval vessel to exist is carried out to the detection on single naval vessel according to the itti visual attention model improved.Gist feature see existing document " Drivingme Around the Bend:Learning to Drive from Visual Gist ", can utilize biological vision feature calculation Gist feature.Embodiment flow process can adopt computer software technology to realize automatically running, and as shown in Figure 1, specifically comprises the following steps:
Step 1, gathers the data of complicated sea remote sensing image.
During concrete enforcement, adopting training data to cover should be many as much as possible, to tackle detecting step below.Embodiment gathers the complicated sea remote sensing image data of different phase, different sensors, different scale.
Step 2, carries out partitioning pretreatment to complicated sea remote sensing image, makes sample data, obtains sample image section and detects image slice.
Embodiment, to remote sensing image block process, is by equal sized by original jumbo image block, and the image slice of N × N size, namely each section comprises N × N number of pixel.
After the complicated sea remote sensing image data piecemeal of the different phases of collection, different sensors, different scale, training sample can be selected from gained image slice, open cut into slices (naval vessel) and S of positive sample image by handmarking S and open negative sample image slice (non-naval vessel), the common training set forming sample image and cut into slices.Or directly can adopt the sample image section marked in advance.Other image slice can as detection image slice to be sorted.General is all using view picture complicated sea remote sensing image as image to be detected, and after image block to be detected, all sub-image blocks of gained are as detection image slice to be sorted.
Wherein, the concrete value of N and S according to circumstances can be set voluntarily by those skilled in the art.
Step 3, extracts notable feature and the Gist feature of sample image section and detection image slice.
In embodiment, notable feature extraction is carried out to arbitrary sample image section and detection image slice, specifically:
Image slice is decomposed on color, brightness, 3, direction feature passage, color, brightness passage use this pyramid representation of 9 floor heights respectively, generate the characteristic pattern on color characteristic passage and the characteristic pattern on brightness passage respectively by the operation of central peripheral difference; Direction character passage Gabor filtering obtains 0 °, 45 °, 90 °, 9 layers of direction pyramid in 135 ° of 4 directions, generates the characteristic pattern on each direction respectively by the operation of central peripheral difference.Wherein, characteristic pattern on color characteristic passage comprises 12 color characteristic subgraphs and (divides two groups, one group is red green color contrast, one group is champac color contrast, often organize each 6), characteristic pattern on brightness passage comprises 6 brightness subgraphs (9 layers of pyramid become 6 after the operation of central peripheral difference), characteristic pattern on each direction of direction character passage comprises 6 direction character subgraphs (9 layers of pyramid become 6 after the operation of central peripheral difference), 24 feature subgraphs altogether on 4 directions.42 feature subgraphs, i.e. 42=12+6+4 × 6 altogether.Then merged by local nonlinearity characteristic pattern is fused to color, brightness, direction significantly scheme, wherein 12 color characteristic subgraphs are fused to 1 color and significantly scheme, 6 brightness subgraphs are fused to 1 brightness significantly schemes, totally 24 direction character subgraphs on 4 directions are fused to 1 direction significantly schemes, and significantly schemes for totally 3.Finally divide other average, standard deviation, Local modulus maxima, Local modulus maxima spacing with each remarkable figure of image slice, the notable feature significantly schemed point 12 dimensional vectors that other 4 values are formed represent image slice with 3.The average of image, standard deviation, Local modulus maxima, Local modulus maxima spacing is specifically asked for, the operation of central peripheral difference and local non-linear fusion are prior art, can see document " L.Itti, C.Koch, E.Niebur, AModel of Saliency-Based Visual Attention for Rapid Scene Analysis, IEEE Transactions on PatternAnalysis and Machine Intelligence, Vol.20, No.11, pp.1254-1259, Nov 1998. ", " L.Itti, C.Koch, Comparison of Feature Combination Strategies for Saliency-Based Visual Attention Systems, In:Proc.SPIE Human Vision and Electronic Imaging IV (HVEI'99), San Jose, CA, Vol.3644, pp.473-82, Bellingham, WA:SPIE Press, Jan 1999. "
In embodiment, Gist feature extraction is carried out to arbitrary sample image section and detection image slice, specifically:
Image slice is decomposed on color, brightness, 3, direction feature passage, color, brightness passage use this pyramid representation of 9 floor heights respectively, generate the characteristic pattern on color characteristic passage and the characteristic pattern on brightness passage respectively by the operation of central peripheral difference; Direction character passage Gabor filtering obtains 0 °, 45 °, 90 °, 4 layers of direction pyramid in 135 ° of 4 directions, the image (i.e. original image obtain divided by 20,21,22,23 respectively 4 tomographic images) getting 4 yardsticks of each direction pyramid, as the direction character subgraph of respective direction, obtains the characteristic pattern on 4 directions.Characteristic pattern on color characteristic passage comprises 12 color characteristic subgraphs, and the characteristic pattern on brightness passage comprises 6 brightness subgraphs, and the characteristic pattern on each direction of direction character passage comprises 4 direction character subgraphs.34 feature subgraphs, i.e. 34=6+12+4 × 4 altogether.4 × 4 sub-grids are divided into often opening feature subgraph, ask the average of 16 sub-grids respectively, average (being exactly as its average of overall calculation by 4 sub-grids) after 4, upper left corner sub-grid, 4, upper right corner sub-grid, 4, lower left corner sub-grid, 4, upper left corner sub-grid, 4, upper left corner sub-grid merge respectively and the average of whole feature subgraph, obtain the vector of one 21 dimension, finally obtain the Gist proper vector of 34 × 21=714 dimension.
During concrete enforcement, the gaussian pyramid number of plies according to circumstances can be set voluntarily by those skilled in the art, and the number of plies is more, then the feature dimensions extracted is also higher.
Step 4, trains sample image section according to step 3 gained notable feature and Gist feature, obtains training pattern.
According to the notable feature of sample image section, and belong to the section of positive sample image or negative sample image slice, can train.Concrete training realization can adopt SVM training aids of the prior art, can obtain training pattern after training to all sample image sections.
Whether step 5, according to step 4 gained training pattern, judge to detect in image slice containing naval vessel by SVM classifier.For arbitrary detection image slice, if judged result is for containing, entering step 6, if judged result is not for contain, terminating the process this being detected to image slice.
Training pattern and predicting the outcome all is realized by existing SVM method, and such as SVM kernel function is with existing RBF kernel function (radial basis function), and it will not go into details in the present invention.
Step 6, finds the single naval vessel detecting image slice based on itti visual attention model, obtain ship target.
Itti visual attention model is implemented as prior art, and for improving for the purpose of Detection accuracy, embodiment proposes to improve itti visual attention model further, finds the single naval vessel detecting image slice based on the itti visual attention model improved.The itti visual attention model of described improvement adds textural characteristics on the basis of original itti model, extend its characteristic range, before the feature subgraph of different scale merges, carry out a square of stretching computing simultaneously, strengthen the difference in remarkable district and non-significant district, outstanding Ship Target.
The step of embodiment is as follows:
Step 6.1, carries out low-level visual features extraction to detection image slice, comprises brightness, color, direction and textural characteristics.Original itti visual attention model extracts brightness, color, direction respectively as low-level visual features, embodiment adds texture feature extraction, texture feature extraction is realized by existing discrete square converter technique, see document " VD.Gesu; C.Valent; L.Strinati.Local operators to detect regions of interest [J] .Pattern Recognition Letter, 1997,18 (11-13): 1077-1081. "
In embodiment, color, brightness passage use this pyramid representation of 9 floor heights respectively, generate the characteristic pattern on color characteristic passage and the characteristic pattern on brightness passage respectively by the operation of central peripheral difference; Direction character passage Gabor filtering obtains 0 °, 45 °, 90 °, 9 layers of direction pyramid in 135 ° of 4 directions, is generated the characteristic pattern on each direction by the operation of central peripheral difference.Wherein, characteristic pattern on color characteristic passage comprises 12 color characteristic subgraphs, characteristic pattern on brightness passage comprises 6 brightness subgraphs, and the characteristic pattern on each direction of direction character passage comprises 6 direction character subgraphs, 24 feature subgraphs altogether on 4 directions.42 feature subgraphs, i.e. 42=12+6+4 × 6 altogether.Then using the feature subgraph of all different scales as input, merged by local nonlinearity characteristic pattern is fused to color, brightness, direction significantly scheme, wherein 12 color characteristic subgraphs are fused to 1 color and significantly scheme, 6 brightness subgraphs are fused to 1 brightness significantly schemes, totally 24 direction character subgraphs on 4 directions are fused into 1 direction significantly schemes, and significantly schemes for totally 3.
Characteristic pattern on the textural characteristics passage adopting discrete square to convert to obtain comprises 3 textural characteristics subgraphs, 3 textural characteristics subgraphs are fused to 1 texture significantly scheme, 3 characteristic remarkable pictures before adding, altogether significantly scheme for 4, they are fused to again 1 entirety significantly to scheme, comprehensive all features.
Step 6.3, Ship Target Detection, the remarkable figure obtained in step 6.2 detects the single Ship Target of image slice by inhibition of return technology for detection.Existing inhibition of return technology can it will not go into details in the present invention see " L.Itti; C.Koch; E.Niebur; A Model ofSaliency-Based Visual Attention for Rapid Scene Analysis; IEEE Transactions on Pattern Analysisand Machine Intelligence, Vol.20, No.11; pp.1254-1259, Nov 1998. ".
Specific embodiment described herein is only to the explanation for example of the present invention's spirit.Those skilled in the art can make various amendment or supplement or adopt similar mode to substitute to described specific embodiment, but can't depart from spirit of the present invention or surmount the scope that appended claims defines.
Claims (2)
1., based on a complicated sea remote sensing image Ship Detection for Gist feature learning, it is characterized in that, comprise the following steps:
Step 1, gathers the data of complicated sea remote sensing image;
Step 2, carries out partitioning pretreatment to complicated sea remote sensing image, obtains sample image section and detects image slice;
Step 3, extracts notable feature and the Gist feature of the section of arbitrary sample image and detection image slice;
Carry out notable feature extraction to arbitrary sample image section and detection image slice in step 3, implementation is as follows,
Arbitrary sample image section or detection image slice are decomposed on color, brightness, these 3 feature passages of direction, color, brightness passage represent with gaussian pyramid respectively, generate the characteristic pattern on color characteristic passage and the characteristic pattern on brightness passage by the operation of central peripheral difference; Direction character passage Gabor filtering obtains 0 °, 45 °, 90 °, the direction pyramid in 135 ° of 4 directions, is generated the characteristic pattern on each direction by the operation of central peripheral difference;
Merged by local nonlinearity and the feature subgraph of each characteristic pattern is fused to color is significantly schemed, brightness is significantly schemed and direction is significantly schemed, finally represent the section of arbitrary sample image by the average of each remarkable figure, standard deviation, Local modulus maxima and Local modulus maxima spacing or detect the notable feature of image slice;
Carry out described Gist feature extraction to arbitrary sample image section and detection image slice in step 3, implementation is as follows,
Arbitrary sample image section or detection image slice are decomposed on color, brightness, 3, direction feature passage, color, brightness passage use this pyramid representation of 9 floor heights respectively, generate the characteristic pattern on color characteristic passage and the characteristic pattern on brightness passage by the operation of central peripheral difference; Direction character passage Gabor filtering obtains 0 °, 45 °, 90 °, and the direction pyramid in 135 ° of 4 directions obtains the characteristic pattern on the passage of each direction;
The sub-grid that feature subgraph is divided into 4 × 4 sizes is often opened to each characteristic pattern, average after asking the average of 16 sub-grids, 4, upper left corner sub-grid, 4, upper right corner sub-grid, 4, lower left corner sub-grid, 4, lower right corner subnet to merge respectively respectively and the average of whole feature subgraph, finally represent Gist feature with acquired results;
Step 4, trains sample image section according to step 3 gained notable feature and Gist feature, obtains training pattern;
By SVM classifier, step 5, according to step 4 gained training pattern, judges that whether arbitrary detection image slice is containing naval vessel, is enter step 5, otherwise terminates the detection to this detection image slice;
Step 6, finds the single naval vessel detecting image slice based on itti visual attention model, obtain ship target; Comprise the following steps,
Step 6.1, carries out low-level visual features extraction to detection image slice, comprises brightness, color, direction and textural characteristics;
Step 6.2, remarkable figure generates, comprise step 6.1 is obtained brightness, color, direction, texture 4 features by central peripheral difference operation generate each feature subgraph, then brightness is significantly schemed, color is significantly schemed, direction is significantly schemed, texture is significantly schemed to merge generation by local nonlinearity fusion method, finally merges by local nonlinearity fusion method and generates overall remarkable figure; In fusion generation, brightness is significantly schemed, color is significantly schemed, direction is significantly schemed, texture carries out square operation to each pixel value of individual features subgraph before significantly scheming;
Step 6.3, Ship Target Detection, comprises the single naval vessel remarkable figure obtained in step 6.2 being detected to image slice by inhibition of return technology for detection, obtains ship target.
2. according to claim 1 based on the complicated sea remote sensing image Ship Detection of Gist feature learning, it is characterized in that: the remote sensing image data selecting to derive from different phase, different sensors, different scale when gathering the data of complicated sea remote sensing image, after carrying out partitioning pretreatment to complicated sea remote sensing image, the positive sample image of mark containing naval vessel and the negative sample image not containing naval vessel are cut into slices as sample image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210339791.1A CN102867196B (en) | 2012-09-13 | 2012-09-13 | Based on the complicated sea remote sensing image Ship Detection of Gist feature learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210339791.1A CN102867196B (en) | 2012-09-13 | 2012-09-13 | Based on the complicated sea remote sensing image Ship Detection of Gist feature learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102867196A CN102867196A (en) | 2013-01-09 |
| CN102867196B true CN102867196B (en) | 2015-10-21 |
Family
ID=47446060
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210339791.1A Active CN102867196B (en) | 2012-09-13 | 2012-09-13 | Based on the complicated sea remote sensing image Ship Detection of Gist feature learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102867196B (en) |
Families Citing this family (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103177608A (en) * | 2013-03-01 | 2013-06-26 | 上海海事大学 | Offshore suspicious ship and ship oil contamination discovering system |
| CN103530628B (en) * | 2013-10-29 | 2016-08-17 | 上海市城市建设设计研究总院 | High-resolution remote sensing image ortho-rectification method based on floating control point |
| CN105335765A (en) * | 2015-10-20 | 2016-02-17 | 北京航天自动控制研究所 | Method for detecting characteristic region matched with SAR |
| CN105243154B (en) * | 2015-10-27 | 2018-08-21 | 武汉大学 | Remote sensing image retrieval method based on notable point feature and sparse own coding and system |
| CN106651937B (en) * | 2016-10-19 | 2019-10-18 | 成都电科智达科技有限公司 | A kind of small drone object detection method based on super-pixel and scene prediction |
| CN106778495A (en) * | 2016-11-21 | 2017-05-31 | 北京航天宏图信息技术股份有限公司 | Ship Detection in remote sensing image under complicated sea background |
| CN107704865A (en) * | 2017-05-09 | 2018-02-16 | 北京航空航天大学 | Fleet Targets Detection based on the extraction of structure forest edge candidate region |
| CN107563303B (en) * | 2017-08-09 | 2020-06-09 | 中国科学院大学 | Robust ship target detection method based on deep learning |
| CN109427055B (en) * | 2017-09-04 | 2022-12-20 | 长春长光精密仪器集团有限公司 | Remote sensing image sea surface ship detection method based on visual attention mechanism and information entropy |
| CN109558771B (en) * | 2017-09-26 | 2023-06-09 | 中电科海洋信息技术研究院有限公司 | Behavior state identification method, device and equipment of marine ship and storage medium |
| CN107977621A (en) * | 2017-11-29 | 2018-05-01 | 淮海工学院 | Shipwreck identification model construction method, device, electronic equipment and storage medium |
| CN108052585B (en) * | 2017-12-11 | 2021-11-23 | 江苏丰华联合科技有限公司 | Method for judging dynamic target in complex environment |
| CN108133184B (en) * | 2017-12-20 | 2020-04-28 | 中国水产科学研究院渔业机械仪器研究所 | Fish identification and vaccine injection method based on fractal theory and BP algorithm |
| CN109800716A (en) * | 2019-01-22 | 2019-05-24 | 华中科技大学 | One kind being based on the pyramidal Oceanic remote sensing image ship detecting method of feature |
| CN110008833B (en) * | 2019-02-27 | 2021-03-26 | 中国科学院半导体研究所 | Target ship detection method based on optical remote sensing image |
| CN109871823B (en) * | 2019-03-11 | 2021-08-31 | 中国电子科技集团公司第五十四研究所 | Satellite image ship detection method combining rotating frame and context information |
| CN110084181B (en) * | 2019-04-24 | 2021-04-20 | 哈尔滨工业大学 | Remote sensing image ship target detection method based on sparse MobileNet V2 network |
| CN111738236B (en) * | 2020-08-14 | 2020-11-20 | 之江实验室 | Image segmentation identification method, device and system of self-adaptive hierarchy |
| CN112766083B (en) * | 2020-12-30 | 2023-10-27 | 中南民族大学 | Remote sensing scene classification method and system based on multi-scale feature fusion |
| CN113256720B (en) * | 2021-06-03 | 2021-09-24 | 浙江大学 | Method for simultaneously detecting SAR image ship and trail thereof |
| CN116310734B (en) * | 2023-04-25 | 2023-12-15 | 慧铁科技股份有限公司 | Fault detection method and system for railway wagon running part based on deep learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102214298A (en) * | 2011-06-20 | 2011-10-12 | 复旦大学 | Method for detecting and identifying airport target by using remote sensing image based on selective visual attention mechanism |
| CN102663348A (en) * | 2012-03-21 | 2012-09-12 | 中国人民解放军国防科学技术大学 | Marine ship detection method in optical remote sensing image |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8243991B2 (en) * | 2008-06-17 | 2012-08-14 | Sri International | Method and apparatus for detecting targets through temporal scene changes |
-
2012
- 2012-09-13 CN CN201210339791.1A patent/CN102867196B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102214298A (en) * | 2011-06-20 | 2011-10-12 | 复旦大学 | Method for detecting and identifying airport target by using remote sensing image based on selective visual attention mechanism |
| CN102663348A (en) * | 2012-03-21 | 2012-09-12 | 中国人民解放军国防科学技术大学 | Marine ship detection method in optical remote sensing image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102867196A (en) | 2013-01-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102867196B (en) | Based on the complicated sea remote sensing image Ship Detection of Gist feature learning | |
| CN110310264B (en) | DCNN-based large-scale target detection method and device | |
| Cheng et al. | FusionNet: Edge aware deep convolutional networks for semantic segmentation of remote sensing harbor images | |
| CN109583369B (en) | Target identification method and device based on target area segmentation network | |
| Nieto-Hidalgo et al. | Two-stage convolutional neural network for ship and spill detection using SLAR images | |
| CN105654091B (en) | Sea-surface target detection method and device | |
| CN109427055B (en) | Remote sensing image sea surface ship detection method based on visual attention mechanism and information entropy | |
| CN107247930A (en) | SAR image object detection method based on CNN and Selective Attention Mechanism | |
| CN106816038A (en) | A kind of inland waters abnormal behaviour ship automatic identification system and method | |
| CN103886285A (en) | Method for detecting optical remote-sensing image ship with assistance of prior geographic information | |
| CN110414509B (en) | Port docking ship detection method based on sea-land segmentation and characteristic pyramid network | |
| Solberg et al. | Automatic detection of oil spills in Envisat, Radarsat and ERS SAR images | |
| CN109117802A (en) | Ship Detection towards large scene high score remote sensing image | |
| CN111079596A (en) | System and method for identifying typical marine artificial target of high-resolution remote sensing image | |
| CN110516605A (en) | Any direction Ship Target Detection method based on cascade neural network | |
| CN110008900B (en) | Method for extracting candidate target from visible light remote sensing image from region to target | |
| CN110647802A (en) | Remote sensing image ship target detection method based on deep learning | |
| CN110659550A (en) | Traffic sign recognition method, traffic sign recognition device, computer equipment and storage medium | |
| Hou et al. | SAR image ship detection based on visual attention model | |
| Zhao et al. | SAR ship detection based on end-to-end morphological feature pyramid network | |
| CN107704865A (en) | Fleet Targets Detection based on the extraction of structure forest edge candidate region | |
| Hou et al. | End-to-end automatic ship detection and recognition in high-resolution Gaofen-3 spaceborne SAR images | |
| He et al. | Ship detection without sea-land segmentation for large-scale high-resolution optical satellite images | |
| Corbane et al. | Fully automated procedure for ship detection using optical satellite imagery | |
| CN105512622A (en) | Visible remote-sensing image sea-land segmentation method based on image segmentation and supervised learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |