CN106951922A - A kind of real-time screening system of astronomic graph picture based on SVMs - Google Patents
A kind of real-time screening system of astronomic graph picture based on SVMs Download PDFInfo
- Publication number
- CN106951922A CN106951922A CN201710157933.5A CN201710157933A CN106951922A CN 106951922 A CN106951922 A CN 106951922A CN 201710157933 A CN201710157933 A CN 201710157933A CN 106951922 A CN106951922 A CN 106951922A
- Authority
- CN
- China
- Prior art keywords
- image
- value
- graph picture
- degree
- real
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to astronomy field, particularly a kind of real-time screening system to astronomic graph picture.Feature extraction is carried out to existing astronomic graph picture first, then astronomic graph picture is classified with SVMs with the grader trained, utilize the optimization function carried under LIBSVM, 7 degree of freedom data to step 2 find optimal loss function and penalty coefficient, build grader, then feature extraction is carried out to current astronomic graph picture, category filter is carried out using the grader built.It provides a kind of quick, accurate screening system in real time to the screening in real time of astronomic graph picture, and automaticity is improved with respect to artificial screening.
Description
Technical field：
The invention belongs to astronomy field, particularly a kind of real-time screening system to astronomic graph picture.
Background technology：
In time domain astronomical observation, telescope general work is toured the heavens under pattern in automatically.In observation, astronomical telescope is solid
Determine the time for exposure, according to the tactful continuous exposure of observation, intensive sampling is carried out to celestial image in time-domain.Due to gathering
The image device of Cheng Zhong, telescope and rear end can will be caused the image of some existing defects to produce by the interference of various noises,
The image of these existing defects can also store transmission in the lump with valid data, and storage and the network money of computer are wasted significantly
Source.Can have a strong impact on automatically extracting for image information additionally, there are the image of defect, cause wrong report and false-alarm, be unfavorable for for
The instant extraction of celestial body time-domain information.Therefore, research how automatic identification and screening existing defects view data for time domain
Astronomical research will be very important.
At present, the method that domestic and international low frame rate observatory uses artificial screening mostly.Although the complexity of artificial screening system
Degree is low, but automaticity is low.
The content of the invention
The technical problems to be solved by the invention are：How automatic identification and screening existing defects view data.
The technical solution adopted in the present invention is：A kind of real-time screening system of astronomic graph picture based on SVMs, is pressed
Carried out according to the steps
Step one：Existing picture under telescope is classified, one group of optimal character subset picture is selected and adds
Label, the view data to optimal character subset picture is normalized, the gradation of image of optimal character subset picture
It is worth between boil down to 0-255；
Lack Step 2: extracting the signal to noise ratio snr of the image of optimal character subset picture, cosmic ray ROAD, class strip
WLK, the category feature index of cloud cover four are fallen into, image data is converted into a 7 degree of freedom data, i.e. signal to noise ratio snr, cosmic ray
ROAD, class strip defect WLK, energy ASM, contrast C ON, unfavourable balance are away from IDM, entropy ENT 7 degree of freedom data；
Step 3: under SVMs, using the optimization function carried under LIBSVM, to the 7 degree of freedom data of step 2
Optimal loss function and penalty coefficient is found, grader is built；
Step 4: the realtime image data under astronomical telescope is normalized, the gray value of image is compressed
For between 0-255, the signal to noise ratio snr of extraction image, cosmic ray ROAD, class strip defect WLK, the category feature of cloud cover four refer to
Mark, a 7 degree of freedom data, i.e. signal to noise ratio snr, cosmic ray ROAD, class strip defect WLK, energy are converted into by image data
ASM, contrast C ON, unfavourable balance are away from IDM, entropy ENT 7 degree of freedom data；
Step 5: being classified to the 7 degree of freedom data extracted in step 4 with the grader built in step 3.
It is used as a kind of preferred embodiment：The signal to noise ratio snr of image is extracted extracts special using formula S NR=10log (Ps/Pn)
Levy, in formula, Ps is the power spectrum of the signal of image, and here using all local variance maximums of image, Pn makes an uproar for image
The power spectrum of sound, adopts all local variance minimum values of image here, and local variance refers to the side of any 9*9 pixels in image
Difference, is accelerated using GPU.GPU is that the calculating in the image processor of computer, computer is mainly by arithmetical logic module (ALU)
Carry out, GPU contains number with 1,100 or more ALU modules with respect to CPU, can be greatly improved by GPU concurrent operation
Operating rate.Concurrent operation installs CUDA modules in computer first, and (CUDA is a parallel computing platform and compiling based on CPU
Model), it would be desirable to the content for carrying out concurrent operation is converted by the data form of CUDA requirements, is calculated in CUDA,
And return to result of calculation.
It is used as a kind of preferred embodiment：When being extracted to the cosmic ray ROAD of image, accelerated using GPU, cosmic ray
ROAD refers to the α particles of the high energy from outer sky, high energy proton, and it has the characteristic of random distribution, but its gray value is obvious
The gray value of projecting pixel.The detection method of the cosmic ray used at present mainly has histogram, Laplce, ROAD
Omnipotent algorithm, the method to these three identification cosmic rays is compared, and the advantage of wherein ROAD algorithms is that algorithm complex is low
It can almost identify all cosmic rays again.But algorithm needs to travel through picture in its entirety, traditional single CPU serially runs reality
Existing efficiency is low, and GPU is employed herein and carries out concurrent operation to improve the speed of service, is made here using the maximum of ROAD in image
Judge to whether there is cosmic ray in image for an evaluation index.
It is used as a kind of preferred embodiment：Gray value in sciagraphy, i.e. image is used to be higher than threshold when class strip defect characteristic is extracted
The pixel of value, is projected in 0 °, 45 °, 90 °, 135 ° of four directions, passes through the ratio of the maxima and minima to four projections
Whether value is more than 2 to judge whether class strip defect, and threshold value is multiplied by 0.9 for gray value maximum in image.Class strip lacks
Fall into reduction picture quality, increase image procossing difficulty.
It is used as a kind of preferred embodiment：The method of the texture recognition used during cloud cover feature extraction, i.e., by image
Overall energy ASM, contrast C ON, comprehensive function of the unfavourable balance away from IDM, tetra- parameters of entropy ENT determine whether that cloud layer hides
Gear.
It is used as a kind of preferred embodiment：The comprehensive function of energy ASM, contrast C ON, unfavourable balance away from IDM, tetra- parameters of entropy ENT
To judge to be specially：If f (x, y) is a width two-dimensional digital image, its size is M*N, and its grey level is Ng, then meets it empty
Between relation gray level co-occurrence matrixes for g (i, j)=# (x1, y1), (x2, y2) ∈ M*N | f (x1, y1)=i, f (x2, y2)=
J }, # { x } represents the element number in set x in formula, if distance is d between (x1, y1) and (x2, y2), both and abscissa line
The gray value that the value that angle is the element g (i, j) in θ, gray level co-occurrence matrixes illustrates one of pixel in the picture is i,
The gray value of one other pixel is j, and neighbor distance is d, the number of times that deflection occurs for θ two such pixel, not
Equidirectional gray level co-occurrence matrixes influence very little to the textural characteristics of image, from d=1, θ=0.
Energy ASM, namely each element quadratic sum, be that the measurement to images flat degree its value is
To different types of picture, the characteristics of gray level co-occurrence matrixes have different：Such as to Continuous Gray Scale value image, value is concentrated on diagonally
Line；To the image of structuring, value concentrates on the cornerwise position of deviation；To the image with critical noisy, gray level co-occurrence matrixes
Distribution value than more uniform.If all values in g are uniform, ASM value is smaller；If some opposite values are big other values compared with
Small, then ASM value is larger.
Contrast C ON is the situation for reflecting image local grey scale change, and its value is
Higher value, i.e. brightness of image value changes are have if the deviation from cornerwise member quickly, then CON has larger value, this is also complied with
The definition of contrast.Gray scale difference is the big pixel of contrast to more, and this value is bigger.Away from diagonal in the raw matrix of gray scale public affairs
Element value it is bigger, CON is bigger.
Unfavourable balance square IDM reflect image texture homogeney, measurement image texture localized variation number, its value isIf gray level co-occurrence matrixes diagonal element have higher value, IDM will take larger value.
Entropy is the measurement for the information content that image has, and is the measurement of a randomness, and it illustrates texture in image
Non-uniform degree or complexity, its value isWhen there being all elements in co-occurrence matrix
When all values are almost equal in maximum randomness, space co-occurrence matrix, in co-occurrence matrix during element dispersed distribution, entropy is larger.
The beneficial effects of the invention are as follows：Feature extraction is carried out to astronomic graph picture first, then trained with SVMs
Good grader is classified to astronomic graph picture.It provides a kind of quick, accurate sieve in real time to the screening in real time of astronomic graph picture
System is selected, automaticity is improved with respect to artificial screening.
Embodiment
Experimental data：The image data for testing the big visual field Schmidt telescopes selected is used as initial data, picture size
For 1024*1024.
Experimental situation：It is the Core i7-5820k that Intel Company produces that experiment, which uses CPU, and dominant frequency is 3.3GHz,
15MB three-level caching.GPU is the GT 610 that NVIDIA companies produce, and possesses 48 stream multiprocessor (Streaming
Multiprocessor, SM), each SM resides thread up to 1024, and computing capability is 2.0, with 4GB video memorys.Program is compiled
Environment is the Python2.7 under ubuntu16.04, and the version that CUDA is used is CUDA parts use in CUDA8.0, program
Python and C shuffling.
Step one：The 20000 width image datas that big visual field Schmidt telescopes have been gathered are chosen, 1000 width are randomly selected
Picture is repeatedly extracted as one group of character subset, to several groups of character subsets, chooses optimal character subset.In character subset
Can there are some signal to noise ratios (SNR), cosmic ray (ROAD), class strip defect (WLK), cloud cover (energy ASM, contrast
CON, unfavourable balance are away from IDM, entropy ENT) four category features are apparent or very unconspicuous picture, and the selection category feature of character subset four is very bright
It is aobvious or very unconspicuous that picture content is minimum and four class defective datas are all present as optimal character subset.Spy
Levy subset image data and be divided into how much two classes (are classified, a class is that content is less, a class according to four class defective data contents
Content it is larger), respectively give label 0,1；Image data is normalized, is 0- by the gray compression of image
255 value.
Step 2：The signal to noise ratio (SNR) of extraction image, cosmic ray (ROAD), class strip defect (WLK), cloud cover
(energy ASM, contrast C ON, unfavourable balance are away from IDM, entropy ENT) four category feature indexs, and image data is converted into a 7 degree of freedom number
According to (i.e. SNR, ROAD, WLK, ASM, CON, IDM, ENT).
Step 3：The data for the tape label handled well to step 2, under SVMs, utilize LIBSVM (LIBSVM
It is one of the exploitation design such as Taiwan Univ.'s woods intelligence benevolence (Lin Chih-Jen) professor simple, easy to use and fast and effectively prop up
Hold vector machine (SVM) pattern-recognition and the software kit returned, can be to use as long as installing the software kit on computers) under carry
Optimization function find optimal loss function and penalty coefficient so that the accuracy of classification is up to more than 90%, use
Optimal penalty coefficient and loss function build grader.
Step 4：Real time data under big visual field Schmidt telescopes is normalized, the noise of image is extracted
Converted than (SNR), cosmic ray (ROAD), class strip defect (WLK), the category feature index of cloud cover four, and by image data
For a 7 degree of freedom data.The dimension in space is characterized shown in following form, the concrete meaning of each formula is with made above
Explanation.
The dimension of the feature space of table 1
Step 5：The characteristic extracted in step 4 is classified with the grader built in step 2, by picture
In the grader that data input is built up.
Wherein, in the image characteristics extraction in step 2 and step 4, the two parameters of signal to noise ratio and cosmic ray are adopted
With the concurrent operation for the CUDA for utilizing GPU, arithmetic speed is improved so that the system can realize the effect handled in real time.GPU
It is the processor of the video card generally used at present, GPU is compared with CPU to be more suitable for handling substantial amounts of parallel data, and GPU has powerful
Floating-point operation, concurrent operation ability, in the feature extraction of image extract SNR and ROAD when to image use circulation time
Go through, and par wise irrelevance between any computing twice, so being adapted to carry out concurrent operation using CUDA.CUDA is video card manufacturer
The computing platform that NVIDIA is released, it is used based on C language, and developer can use C language come the program of writing.
Here main part use Python to be write, wherein CUDA part be use C language.Picture
2-D data for 1024*1024 is and incoming by data by CPU, it is necessary to 2-D data to be changed into 1*1024^2 one-dimensional data
CUDA, carries out concurrent operation in CUDA, and the result outflow after calculating is being carried out into follow-up work to CPU by CUDA.
Claims (6)
1. a kind of real-time screening system of astronomic graph picture based on SVMs, it is characterised in that：Carried out according to the steps
Step one：Existing picture under telescope is classified, one group of character subset of selection is used as optimal character subset figure
Piece simultaneously adds label, and the view data to optimal character subset picture is normalized, optimal character subset picture
Between image intensity value boil down to 0-255；
Step 2: extracting the signal to noise ratio snr of the image of optimal character subset picture, cosmic ray ROAD, class strip defect
WLK, the category feature index of cloud cover four, a 7 degree of freedom data, i.e. signal to noise ratio snr, cosmic ray are converted into by image data
ROAD, class strip defect WLK, energy ASM, contrast C ON, unfavourable balance are away from IDM, entropy ENT 7 degree of freedom data；
Step 3: under SVMs, using the optimization function carried under LIBSVM, the 7 degree of freedom data to step 2 are found
Optimal loss function and penalty coefficient, builds grader；
Step 4: the realtime image data under astronomical telescope is normalized, the gray value boil down to 0- of image
Between 255, the signal to noise ratio snr of extraction image, cosmic ray ROAD, class strip defect WLK, the category feature index of cloud cover four,
By image data be converted into a 7 degree of freedom data, i.e. signal to noise ratio snr, cosmic ray ROAD, class strip defect WLK, energy ASM,
Contrast C ON, unfavourable balance are away from IDM, entropy ENT 7 degree of freedom data；
Step 5: being classified to the 7 degree of freedom data extracted in step 4 with the grader built in step 3.
2. a kind of real-time screening system of astronomic graph picture based on SVMs according to claim 1, it is characterised in that：
The signal to noise ratio snr of image is extracted to be extracted in feature, formula using formula S NR=10log (Ps/Pn), and Ps is all parts of image
Variance maximum, Pn is all local variance minimum values of image, and local variance refers to the variance of any 9*9 pixels in image,
Accelerated using GPU.
3. a kind of real-time screening system of astronomic graph picture based on SVMs according to claim 1, it is characterised in that：
When being extracted to the cosmic ray ROAD of image, accelerated using GPU.
4. a kind of real-time screening system of astronomic graph picture based on SVMs according to claim 1, it is characterised in that：
It is higher than the pixel of threshold value when class strip defect characteristic is extracted using gray value in sciagraphy, i.e. image, at 0 °, 45 °, 90 °, 135 °
Four direction is projected, and judges whether to deposit by the way that whether the ratio of the maxima and minima to four projections is more than 1.5
In class strip defect, threshold value is multiplied by 0.9 for gray value maximum in image.
5. a kind of real-time screening system of astronomic graph picture based on SVMs according to claim 1, it is characterised in that：
The method of the texture recognition used during cloud cover feature extraction, i.e., by the overall energy ASM of image, contrast C ON, inverse
Gap IDM, the comprehensive function of entropy ENT tetra- parameters determine whether cloud cover.
6. a kind of real-time screening system of astronomic graph picture based on SVMs according to claim 5, it is characterised in that：
The comprehensive function of energy ASM, contrast C ON, unfavourable balance away from IDM, tetra- parameters of entropy ENT judges to be specially：If f (x, y) is one
Width two-dimensional digital image, its size is M*N pixels, and its grey level is Ng, then meets the gray level co-occurrence matrixes of its spatial relationship
For g (i, j)=# { (x1, y1), (x2, y2) ∈ M*N | f (x1, y1)=i, f (x2, y2)=j }, if (x1, y1) and (x2, y2)
Between distance be d, both angles with abscissa line are that the value of the element g (i, j) in θ, gray level co-occurrence matrixes is illustrated in image
In the gray value of one of pixel be i, the gray value of one other pixel is j, and neighbor distance is d, and deflection is this of θ
The textural characteristics of image are influenceed very little, from d=by the number of times that two pixels of sample occur in the gray level co-occurrence matrixes of different directions
1, θ=0, x1 and y1 are the transverse and longitudinal coordinate value of any one element in two-dimensional digital image, x2 and y2 be in two-dimensional digital image not
It is same as the transverse and longitudinal coordinate value of any one element of element (x1, y1)；
Energy ASM, namely each element quadratic sum, be that the measurement to images flat degree its value is
Contrast C ON is the situation for reflecting image local grey scale change, and its value is
Unfavourable balance square IDM reflect image texture homogeney, measurement image texture localized variation number, its value is
Entropy is the measurement for the information content that image has, and is the measurement of a randomness, it illustrate texture in image it is non-
Even degree or complexity, its value is
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710157933.5A CN106951922A (en) | 2017-03-16 | 2017-03-16 | A kind of real-time screening system of astronomic graph picture based on SVMs |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710157933.5A CN106951922A (en) | 2017-03-16 | 2017-03-16 | A kind of real-time screening system of astronomic graph picture based on SVMs |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN106951922A true CN106951922A (en) | 2017-07-14 |
Family
ID=59472572
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710157933.5A Pending CN106951922A (en) | 2017-03-16 | 2017-03-16 | A kind of real-time screening system of astronomic graph picture based on SVMs |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106951922A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102022A (en) * | 2018-08-13 | 2018-12-28 | 淮北师范大学 | A kind of light curve classification method based on support vector machines |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116744A (en) * | 2013-02-05 | 2013-05-22 | 浙江工业大学 | Fake fingerprint detection method based on markov random field (MRF) and support vector machine-k nearest neighbor (SVM-KNN) classification |
| CN103500449A (en) * | 2013-09-26 | 2014-01-08 | 西安空间无线电技术研究所 | Satellite visible light remote sensing image cloud detection method |
| CN103714148A (en) * | 2013-12-24 | 2014-04-09 | 西安电子科技大学 | SAR image search method based on sparse coding classification |
| CN103942540A (en) * | 2014-04-10 | 2014-07-23 | 杭州景联文科技有限公司 | False fingerprint detection algorithm based on curvelet texture analysis and SVM-KNN classification |
| CN104000619A (en) * | 2014-06-16 | 2014-08-27 | 彭文献 | Thyroid CT image computer-aided diagnosis system and method |
| CN104866862A (en) * | 2015-04-27 | 2015-08-26 | 中南大学 | Strip steel surface area type defect identification and classification method |
| CN105447504A (en) * | 2015-11-06 | 2016-03-30 | 中国科学院计算技术研究所 | Traffic mode behavior recognition method and corresponding recognition model construction method |
-
2017
- 2017-03-16 CN CN201710157933.5A patent/CN106951922A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103116744A (en) * | 2013-02-05 | 2013-05-22 | 浙江工业大学 | Fake fingerprint detection method based on markov random field (MRF) and support vector machine-k nearest neighbor (SVM-KNN) classification |
| CN103500449A (en) * | 2013-09-26 | 2014-01-08 | 西安空间无线电技术研究所 | Satellite visible light remote sensing image cloud detection method |
| CN103714148A (en) * | 2013-12-24 | 2014-04-09 | 西安电子科技大学 | SAR image search method based on sparse coding classification |
| CN103942540A (en) * | 2014-04-10 | 2014-07-23 | 杭州景联文科技有限公司 | False fingerprint detection algorithm based on curvelet texture analysis and SVM-KNN classification |
| CN104000619A (en) * | 2014-06-16 | 2014-08-27 | 彭文献 | Thyroid CT image computer-aided diagnosis system and method |
| CN104866862A (en) * | 2015-04-27 | 2015-08-26 | 中南大学 | Strip steel surface area type defect identification and classification method |
| CN105447504A (en) * | 2015-11-06 | 2016-03-30 | 中国科学院计算技术研究所 | Traffic mode behavior recognition method and corresponding recognition model construction method |
Non-Patent Citations (6)
| Title |
|---|
| QUAN LIAO 等: "GPU Accelerated Support Vector Machines for Mining High-Throughput Screening Data", 《JOURNAL OF CHEMICAL INFORMATION AND MODELING》 * |
| ZHIXIAN MA 等: "X-ray Astronomical Point Sources Recognition Using Granular Binary-tree SVM", 《ICSP2016》 * |
| 刘婷婷 等: "消除CCD图像中宇宙射线的算法的比较", 《天文研究与技术》 * |
| 赵敏 等: "一种新的可见光遥感图像云判别算法", 《遥感技术与应用》 * |
| 邱浩阳 等: "基于支持向量机的导航星筛选方法", 《兵工学报》 * |
| 马志贤 等: "基于SVM的X射线天文图像点源探测算法", 《上海师范大学学报(自然科学版)》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109102022A (en) * | 2018-08-13 | 2018-12-28 | 淮北师范大学 | A kind of light curve classification method based on support vector machines |
| CN109102022B (en) * | 2018-08-13 | 2020-07-03 | 淮北师范大学 | Light variation curve classification method based on support vector machine |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10402697B2 (en) | Fusing multilayer and multimodal deep neural networks for video classification | |
| Han et al. | Underwater image processing and object detection based on deep CNN method | |
| US20180101768A1 (en) | Temporal ensembling for semi-supervised learning | |
| Fu et al. | Learning to detect specular highlights from real-world images | |
| US11748887B2 (en) | Segmentation using an unsupervised neural network training technique | |
| Manfredi et al. | Shift equivariance in object detection | |
| O’Keeffe et al. | A benchmark data set and evaluation of deep learning architectures for ball detection in the RoboCup SPL | |
| CN108648213A (en) | A kind of implementation method of KCF track algorithms on TMS320C6657 | |
| Hu et al. | Mask-guided cycle-GAN for specular highlight removal | |
| Chaitra et al. | Deep-CNNTL: text localization from natural scene images using deep convolution neural network with transfer learning | |
| Pan et al. | FPGA-accelerated textured surface defect segmentation based on complete period Fourier reconstruction | |
| Moya-Sánchez et al. | A trainable monogenic convnet layer robust in front of large contrast changes in image classification | |
| Singh et al. | Parallel implementation of Otsu’s binarization approach on GPU | |
| CN106951922A (en) | A kind of real-time screening system of astronomic graph picture based on SVMs | |
| Akagic et al. | Superpixel accelerator for computer vision applications on arria 10 soc | |
| Patlolla et al. | Accelerating satellite image based large-scale settlement detection with GPU | |
| van Werkhoven et al. | A Jungle Computing approach to common image source identification in large collections of images | |
| CN116309168A (en) | Low-illumination image enhancement method and system for parallel hybrid attention progressive fusion | |
| US20220129755A1 (en) | Incorporating a ternary matrix into a neural network | |
| Li et al. | Toward a Deeper understanding: RetNet viewed through convolution | |
| Bandara et al. | Nature inspired dimensional reduction technique for fast and invariant visual feature extraction | |
| Arora et al. | GPU approach for handwritten devanagari document binarization | |
| Xiaojuan et al. | An improved BP neural network for wastewater bacteria recognition based on microscopic image analysis | |
| Qin et al. | YOLOV3 traffic sign recognition and detection based on FPN improvement | |
| Gupta et al. | Ultramnist classification: A benchmark to train cnns for very large images |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20170714 |