CN109522889A - Hydrological ruler water level identification and estimation method based on image analysis - Google Patents
Hydrological ruler water level identification and estimation method based on image analysis Download PDFInfo
- Publication number
- CN109522889A CN109522889A CN201811020954.3A CN201811020954A CN109522889A CN 109522889 A CN109522889 A CN 109522889A CN 201811020954 A CN201811020954 A CN 201811020954A CN 109522889 A CN109522889 A CN 109522889A
- Authority
- CN
- China
- Prior art keywords
- ruler
- hydrology
- character
- image
- size
- 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
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- 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
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/14—Image acquisition
- G06V30/146—Aligning or centring of the image pick-up or image-field
- G06V30/1475—Inclination or skew detection or correction of characters or of image to be recognised
- G06V30/1478—Inclination or skew detection or correction of characters or of image to be recognised of characters or characters lines
-
- 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
- G06V2201/02—Recognising information on displays, dials, clocks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Abstract
The invention discloses a hydrological ruler water level identification and estimation method based on image analysis, which comprises the following steps of: firstly, realizing initial positioning of a hydrological ruler through an HSV color space, and then adaptively constructing a component diagram to realize accurate positioning of the hydrological ruler; expanding samples by a data enhancement technology, realizing character segmentation on the positioned hydrological ruler image by adopting a fuzzy C clustering method, then constructing a convolutional neural network training data set and identifying complete characters; and sequentially calculating the pixel heights of the complete characters E, establishing a mapping relation between the pixel heights and the actual heights in the quadratic equation dynamic fitting image on the basis of the pixel heights of the complete characters E, and further calculating the actual heights of the incomplete characters E. The invention can realize the recognition and estimation of the water depth in the image containing the hydrological gauge, and is suitable for being integrated at the front end of a terminal, such as a video camera, a camera and the like, to automatically estimate the water level of the hydrological gauge.
Description
Technical field
The present invention relates to the automatic measurement of the depth of water, belong to the measurement method based on image, belong to depth of water automatic measurement field,
Specifically a kind of hydrology ruler water level based on image analysis identifies evaluation method.
Background technique
Hydrologic monitoring influences the subject of the people's livelihood, economic construction as one, establishes the hydrology near more crucial waters
It stands and realizes the measurement of the elements such as water level, flow and flow velocity, on the one hand hydrology routine data is provided, be on the other hand water resource benefit
Guaranteed with, traffic and prevention flood etc. provides decision support data.Traditional monitoring mode is that survey crew is logical
It crosses and hydrology ruler is read and recorded water level.With the rapid development of information technology, more and more automatic measurers
For hydrologic monitoring, measurement accuracy and accuracy are improved while saving labor cost.But automatic measurement sensor
Easily affected by environment or instrument is impaired and influences measurement accuracy, using infrared survey, ultrasonic measurement as the non-cpntact measurement of representative
Method then needs periodically to be calibrated and safeguarded, therefore also will use biography in addition to using sensor measurement water level in hydrometric station
The hydrology ruler measurement water level of system is for comparing and verifying the water level sensors such as electronics hydrology ruler.
Meanwhile image processing techniques is widely used in fields of measurement in recent years, such as instrument automatic reading, river
Stream, the level measuring of open channel and water on urban streets monitoring etc..Water level automatic monitoring technical based on image recognition is to combine
Image capture device, common hydrology ruler realize level measuring using contactless measurement, and cost is lower, is hardly damaged, together
When periodic maintenance, the demand of calibration it is relatively low, the practical valence with higher especially under measurement accuracy scene of less demanding
Value can provide data support for agricultural irrigation, Urban Flood control and traffic trip etc., control band for station data quality
Carry out new way, raising automatically processes level.
Summary of the invention
Goal of the invention: being directed to the above-mentioned prior art, proposes a kind of hydrology ruler water level identification estimation side based on image analysis
Method, for estimating hydrology ruler water level automatically.
A kind of technical solution: hydrology ruler water level identification evaluation method based on image analysis, comprising the following steps:
Step 1: picture of the shooting containing hydrology ruler image realizes hydrology ruler Primary Location by hsv color space first,
Then adaptively building component map realizes that hydrology ruler is accurately positioned；
Step 2: character and number on artificial segmentation hydrology ruler, and enhance technology EDS extended data set by data, then
It is randomly divided into training set and test set；Building convolutional neural networks are simultaneously trained and test；Using the method pair of fuzzy C-means clustering
Hydrology ruler image is clustered and is divided, and the character being partitioned into is identified using trained convolutional neural networks, is identified
As a result as the integer part of the depth of water；
Step 3: calculating the pixels tall for the complete character " E " that step 2 identifies in order, and establish based on this
Quadratic equation is fitted in image corresponding relationship between pixels tall and actual height, calculates incomplete digital or incomplete character " E "
Pixels tall, and fractional part of the actual height of incomplete number or incomplete character " E " as the depth of water is calculated by fit correlation；
Step 4: the depth of water result in image is calculated according to step 2 character recognition and step 3 mapping relations.
The utility model has the advantages that a kind of water gauge water level based on image analysis of the invention identifies evaluation method, hydrology ruler figure is realized
The automatic identification of water level and scale estimation, may be implemented to estimate the depth of water by identifying in the image containing hydrology ruler, fit as in
Conjunction is integrated in the front ends such as terminal, such as video camera, camera, automatic to estimate water gauge water level.
Detailed description of the invention
Fig. 1 is overall procedure schematic diagram of the invention；
Fig. 2 is the partial data image of data enhancing；
Fig. 3 is the convolutional neural networks structure of design；
Fig. 4 is two kinds of examples that step 4 situation divides.
Specific embodiment
Further explanation is done to the present invention with reference to the accompanying drawing.
As shown in Figure 1, a kind of hydrology ruler water level based on image analysis identifies evaluation method, comprising the following steps:
Step 1: picture of the shooting containing hydrology ruler image, and two step hydrology ruler location algorithm processes are established, pass through first
Hydrology ruler Primary Location is realized in hsv color space, and then adaptively building component map realizes that hydrology ruler is accurately positioned.
Step 2, character recognition: character and number on artificial segmentation hydrology ruler, and enhanced based on this by data
Technology EDS extended data set, and Character segmentation, segmentation are carried out to the hydrology ruler image that step 1 positions using the method for fuzzy C-means clustering
The character " E " of the number of complete number, complete character " E " and incompleteness or incompleteness out, then constructs convolutional neural networks
Training dataset simultaneously identifies complete character, and in this, as the integer portion of the depth of water.
Step 3, pixels tall and actual height dynamic mapping: calculating the pixels tall of multiple complete characters " E " in order,
And establish quadratic equation based on this and be fitted in image corresponding relationship between pixels tall and actual height, calculate incomplete number
Or the pixels tall of incomplete character " E ", and fractional part of the actual height as the depth of water is calculated by fit correlation.
Step 4, the depth of water result in image is calculated according to step 2 character recognition and step 3 mapping relations.
Step 1 of the present invention comprising the following specific steps
Step 1-1, hydrology ruler Primary Location: picture of the shooting containing hydrology ruler image, and its color space is turned by RGB
It is changed to HSV, and primarily determines the position of hydrology ruler using Threshold segmentation, Morphological scale-space method；
Step 1-2, hydrology ruler are accurately positioned: being constructed adaptive component map (Component map) and be accurately positioned hydrology ruler；
Hydrology ruler Slant Rectify: step 1-3 carries out Slant Rectify to hydrology ruler using Hough transform, extracts from image
Hydrology ruler image out.
Step 1-1 hydrology ruler Primary Location of the present invention comprising the following specific steps
Color space conversion: original image color space is converted to HSV space by RGB, and carries out two by step 1-1-1
Value processing and filtering；
Morphological scale-space: step 1-1-2 carries out Morphological scale-space to the bianry image of step 1-1-1 to primarily determine water
Literary ruler region；
Step 1-1-3, preliminary hydrology ruler segmentation: being split original image according to the hydrology ruler region of step 1-1-2,
Primary segmentation goes out hydrology ruler image.
Step 1-1-1 color space conversion of the present invention comprising the following specific steps
The conversion of hsv color space: step 1-1-1-1 the color space of original image is converted by rgb space to HSV sky
Between, it can extract color (Hue), three kinds of saturation degree (Saturation), brightness (Value) components；
Threshold segmentation: step 1-1-1-2 determines global threshold using Otsu algorithm (OTSU) for Hue component；
Image binaryzation: step 1-1-1-3 according to the global threshold determined in step 1-1-1-2, carries out Hue component
Binary conversion treatment can tentatively show hydrology ruler, exist simultaneously more noise；
Step 1-1-1-4, median filtering:, can using the bianry image in the method processing step 1-1-1-3 of median filtering
Isolated noise is effectively removed, different median filtering parameters can be used for different resolution:
Image resolution ratio is 6000 × 4000, using 50 × 50 parameter as median filtering；
Image resolution ratio is 2560 × 1920, using 20 × 20 parameter as median filtering；
Image resolution ratio is 1600 × 1200, using 20 × 20 parameter as median filtering；
Image resolution ratio is 1024 × 768, using 20 × 20 parameter as median filtering；
Image resolution ratio is 800 × 600, using 10 × 10 parameter as median filtering；
Image resolution ratio is 640 × 480, using 5 × 5 parameter as median filtering.
Step 1-1-2 Morphological scale-space of the present invention comprising the following specific steps
Step 1-1-2-1, closed operation: the image that step 1-1-1-4 is obtained using the closed operation in morphological method into
Row processing, can restore the connection region of hydrology ruler, square structure can be used in closed operation, and size is adjusted according to pixel difference
It is whole；
Opening operation: step 1-1-2-2 carries out opening operation to the image in step 1-1-2-1 to filter out tiny portion in image
Point, square structure can be used in opening operation, and size is adjusted according to pixel difference；
Step 1-1-2-3 removes small connected domain: connected domain mark is carried out to gained image in step 1-1-2-2, according to connection
Logical area size removes small connected domain, leaves hydrology ruler connected domain.
The preliminary hydrology ruler segmentation of step 1-1-3 of the present invention comprising the following specific steps
Step 1-1-3-1 determines hydrology ruler region: boundary rectangle is done into the hydrology ruler region of step 1-1-2-3, it can be appropriate
Rectangular extent is expanded；
Step 1-1-3-2, hydrology ruler Primary Location: by the bianry image of step 1-1-3-1 and original image carries out and operation, i.e.,
The Primary Location of hydrology ruler can be achieved；
Step 1-1-3-3, the segmentation of hydrology ruler；The hydrology ruler image of step 1-1-3-2 is cut, complicated back is removed
Scape.
Step 1-2 hydrology ruler of the present invention be accurately positioned comprising the following specific steps
Step 1-2-1 constructs component map: defining component map (Component map) according to the different colours of hydrology ruler, such as
Fruit hydrology ruler is red, then defines component map are as follows:
Hydrology ruler is blue, then defines component map are as follows:
Wherein C is component map, and R, G, B is hydrology ruler color component, and ω is the parameter for adjusting color weight, directly affects water
The separable degree of literary ruler, p are component factor, and e is the truth of a matter of natural logrithm；
Step 1-2-2, adaptively gets parms: degree of isolation is calculated to hydrology ruler image using OTSU, with degree of isolation
Color weight parameter ω is adaptively determined for standard, so that ω when degree of isolation maximum is as color weight；
Median filtering: step 1-2-3 using the method for median filtering, eliminates hydrology ruler inverted image, finally realizes hydrology ruler
It is accurately positioned.
Step 1-3 hydrology ruler Slant Rectify of the present invention comprising the following specific steps
Edge detection: step 1-3-1 carries out edge to image obtained by step 1-2-3 using the sobel operator of vertical structure
Detection；
Step 1-3-2, Hough transform: hydrology ruler has certain linear feature, using in Hough transform detection image
Straight line, can detect that hydrology ruler vertical edge, be inclined two straight lines；
Rotational correction: step 1-3-3 is process object with hydrology ruler edge, its tilt angle is calculated, then according to inclination
Angle carries out Slant Rectify to hydrology ruler image.
Step 2 of the present invention comprising the following specific steps
Step 2-1, data set prepare: character and number on artificial segmentation hydrology ruler, and pass through data based on this
Enhancing technology EDS extended data set, is further randomly divided into training set and test set；
Step 2-2, convolutional neural networks training: building convolutional neural networks are simultaneously trained and test；
Step 2-3, character recognition: carrying out character cluster to hydrology ruler image using the method for fuzzy C-means clustering and divide, will
The character being partitioned into is identified using trained convolutional neural networks.
Step 2-1 data set of the present invention prepare comprising the following specific steps
Step 2-1-1, makes sample library: the character on artificial segmentation hydrology ruler image, including number, character " E ", simultaneously
It is labeled, totally 11 class, makes sample library；
Step 2-1-2, as shown in Fig. 2, data enhance: the picture of cutting is hung down using change of scale, random cropping, level
Straight overturning, translation transformation, noise disturbance, affine transformation are handled, and last size is normalized to 40 × 40, as data set；
Step 2-1-3, is randomly assigned: the data set of acquisition being selected 1/5 as test set at random, remaining is as training
Collection.
Step 2-2 convolutional neural networks of the present invention training comprising the following specific steps
Step 2-2-1, as shown in figure 3, network struction: using convolutional neural networks structure, network shares 7 layers, the past
After be followed successively by input layer, C1 convolutional layer, S2 down-sampling layer, C3 convolutional layer, S4 down-sampling layer, full articulamentum and output layer；
Step 2-2-2, network training: input training dataset, and initialize network and start to train；
Step 2-2-3, network test: input test data set tests trained network, and recording accuracy.
Step 2-2-1 network struction of the present invention comprising the following specific steps
Step 2-2-1-1: determine each convolutional layer output characteristic pattern OutFeatureMaps, convolution kernel kernel and partially
Bias is set, output characteristic pattern OutFeatureMaps includes characteristic pattern quantity OutFeatureMaps_Num and characteristic pattern size
OutFeatureMaps_Size；Convolution kernel kernel includes convolution kernel size Kernel_Size and convolution nuclear volume
Kernel_Num；The output characteristic pattern quantity OutFeatureMaps_Num phase of the quantity Bias_Num of offset parameter and respective layer
Together.
For convolutional layer Cx, x ∈ { 1,3 }, the output characteristic pattern width of this layer is OutFeatureMaps_Size_Cx,
Value is total by its input feature vector figure (datagram) resolution ratio InFeatureMaps_Size and convolution kernel size Kernel_Size_Cx
With decision, i.e. OutFeatureMaps_Size_Cx=InFeatureMaps_Size-Kernel_Size_Cx+1；
For convolutional layer C1, enabling it export characteristic pattern quantity is OutFeatureMaps_Num_C1=6, exports characteristic pattern
Width is OutFeatureMaps_Size_C1=36；The convolution kernel size Kernel_Size_C1=5 of convolutional layer C1, convolutional layer
The convolution nuclear volume Kernel_Num_C1=6 of C1 initializes convolution kernel numerical value using Xavier method, and initial value isBetween 5 rank matrixes；The quantity Bias_Num_C1=6 of the offset parameter of convolutional layer C1, initial value
It is 0.
For convolutional layer C3, enabling it export characteristic pattern quantity is OutFeatureMaps_Num_C1=12, exports characteristic pattern
Width is OutFeatureMaps_Size_C1=14；The convolution kernel size Kernel_Size_C1=5 of convolutional layer C3, convolutional layer
The convolution nuclear volume Kernel_Num_C1=12 of C3 initializes convolution kernel numerical value using Xavier method, and initial value isBetween 5 rank matrixes；The quantity Bias_Num_C1=12 of the offset parameter of convolutional layer C3, just
Initial value is 0.
Step 2-2-1-2, down-sampling layer building: sample size scale=2 is arranged in down-sampling layer, using average Chi Huafang
The sampling core of down-sampling layer S2, S4 are initialized as by methodThe output characteristic pattern quantity of sample level S2, S4
OutFeatureMaps_Num is equal with its upper layer convolutional layer output characteristic pattern quantity, exports characteristic pattern resolution ratio
OutFeatureMaps_Size is that its upper layer convolutional layer exports characteristic pattern resolution ratio half, it may be assumed that
OutFeatureMaps_Num_S2=6, OutFeatureMaps_Size_S2=18,
OutFeatureMaps_Num_S4=12, OutFeatureMaps_Size_S2=7.
Step 2-2-1-3, full articulamentum construction: constructs full articulamentum, full articulamentum is connected with output layer, weight parameter FW
Initial value isBetween size be 11 × 588 matrix, offset parameter FB initial value be 0 size
For 11 × 1 vector, each data finally export the probability value of all kinds of labels, take label corresponding to maximum probability for output knot
Fruit.
Step 2-2-2 network training of the present invention comprising the following specific steps
Training set pretreatment: training data is concentrated each secondary gray scale diagram data by [0,255] normalizing by step 2-2-2-1
Change between [0,1], and corresponding label matches and is input to the input layer of network；
Step 2-2-2-2, netinit: the learning rate alpha=0.1 of setting network, setting batch size batchsize
=5, set frequency of training iteration=1000；
Step 2-2-2-3 starts to train: using sigmoid function as activation primitive, using secondary cost function conduct
Loss function starts network training, reaches frequency of training iteration deconditioning.
Step 2-2-3 network test of the present invention comprising the following specific steps
Test set pretreatment: test data is concentrated each secondary gray scale diagram data by [0,255] normalizing by step 2-2-3-1
Change between [0,1], being input to trained network and is tested；
Step 2-2-2-2 starts to test: after starting test, outputing test result, carries out with label corresponding to test set
It compares, recording accuracy.
Step 2-3 Character segmentation of the present invention the following steps are included:
Step 2-3-1, character cluster: and hydrology ruler image is clustered using the method for fuzzy C-means clustering, white is carried on the back
Scape and number, character etc. are clustered, to distinguish background and character；
Step 2-3-2, Character segmentation: by complete cluster character be split, after the completion of segmentation, can according to from up to
Under, the arrangement of sequence from left to right, last character is incomplete since water level blocks appearance, therefore last character is defined as
Imperfect character, it may be possible to incomplete character, it is also possible to it is incomplete character " E ", remaining is complete character, including
Complete number and complete character " E "；
Step 2-3-3, character recognition: the complete character that will be partitioned into, including complete number and complete character " E ",
Its size is normalized to 40 × 40, is identified using convolutional neural networks, integer part of the recognition result as the depth of water；And
Imperfect character is calculated using dynamic mapping algorithm, i.e. step 3, acquired results are the fractional part of the depth of water.
Step 3 of the present invention comprising the following specific steps
Step 3-1, pixels tall statistics: in order by the complete character " E " identified in step 2-3-3, i.e., according to
From top to bottom, sequential storage from left to right amounts to m character " E ", and calculates its pixels tall, is denoted as hE1, hE2, hE3...,
hEm；Using the last one incomplete character as m+1, pixels tall hE(m+1)；
Quadratic fit mapping relations: step 3-2 is closed using the mapping between quadratic equation match pixel and actual height
System, by hE1, (hE1+hE2), (hE1+hE2+hE3) ..., (hE1+hE2+hE3+…+hEm) it is used as independent variable x, represent accumulated pixel height
Degree；It corresponds to 50,100,150 ..., 50n represents actual height as dependent variable y, and unit is millimeter mm；To quadratic equation, i.e.,
Y=ax2+ bx+c is fitted, and solves coefficient a, b, c；
Step 3-3, actual height calculate: calculating the actual height that last incomplete character represents, enable x=(hE1+hE2+hE3
+…+hEm+hE(m+1)), bring equation y=ax into2Y is solved in+bx+c, calculating last its actual height of incomplete character is (y-
50m)mm。
Step 4 of the present invention comprising the following specific steps
Step 4-1, as shown in figure 4, situation divides: using different calculations to different situations, be divided into integer portion
Divide to calculate and be calculated with fractional part, the depth of water that complete character recognition can determine that passes through incomplete character as integer part
The depth of water that dynamic mapping calculates is as fractional part；
Step 4-2, integer portion calculate: according to the positional relationship of water level line and hydrology ruler, two kinds of situations can be divided into, and according to
Different situations establishes different rule for read data, and with the left-half of hydrology ruler, i.e., the half comprising number and character " E " is mark
Standard establishes two kinds of situations:
Situation 1, since water level line blocks, water level wire cutting character " E " forms hydrology ruler left-half the last character
Symbol is incomplete character " E ", and for complete character, last character is number, and recognition result n can be primarily determined
Depth of water range is (10n-5,10n) cm, and in this, as integer part.
Situation 2, since water level line blocks, water level wire cutting number forms hydrology ruler right half part last character
For incomplete number, for complete character, last character is character " E ", and penultimate character is number, is known
Other result is n, can primarily determine that depth of water range is (10n, 10n+5) cm, and in this, as integer part.
Step 4-3, the depth of water calculate: calculating fractional part, the reality of the last one incomplete character " E " can be calculated by step 3
Height is h cm, then according to, as a result, arranging the integer part and fractional part of the depth of water, calculating depth of water height is divided to two in step 4-2
Kind situation carries out depth of water calculating:
Situation 1 determines that integer part depth of water range is (10n-5,10n) cm, it is last then to calculate hydrology ruler left-half
The pixels tall of one incomplete character " E ", and its actual height h cm is calculated using dynamic mapping, then last measurement result is
(10n-h)cm。
Situation 2 determines that integer part depth of water range is (10n, 10n+5) cm, it is last then to calculate hydrology ruler right half part
The pixels tall of one incomplete character " E ", and its actual height h cm is calculated using dynamic mapping, then last measurement result is
(10n+5-h)cm。
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (10)
1. a kind of hydrology ruler water level based on image analysis identifies evaluation method, which comprises the following steps:
Step 1: picture of the shooting containing hydrology ruler image realizes hydrology ruler Primary Location by hsv color space first, then
Adaptive building component map realizes that hydrology ruler is accurately positioned；
Step 2: character and number on artificial segmentation hydrology ruler, and enhance technology EDS extended data set by data, then at random
It is divided into training set and test set；Building convolutional neural networks are simultaneously trained and test；Using the method for fuzzy C-means clustering to the hydrology
Ruler image is clustered and is divided, and the character being partitioned into is identified using trained convolutional neural networks, recognition result
Integer part as the depth of water；
Step 3: calculating the pixels tall for the complete character " E " that step 2 identifies in order, and establish based on this secondary
Corresponding relationship between pixels tall and actual height in equation model image calculates the pixel of incomplete number or incomplete character " E "
Highly, and by fit correlation fractional part of the actual height of incomplete number or incomplete character " E " as the depth of water is calculated；
Step 4: the depth of water result in image is calculated according to step 2 character recognition and step 3 mapping relations.
2. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, which is characterized in that
In the step 1, realize that hydrology ruler Primary Location comprises the following specific steps that by hsv color space:
Step 1-1-1: original image color space is converted into HSV space by RGB, and carries out binary conversion treatment and filtering；
Step 1-1-2: Morphological scale-space is carried out to primarily determine hydrology ruler region to the bianry image that step 1-1-1 is obtained；
Step 1-1-3: the hydrology ruler region obtained according to step 1-1-2 is split original image, and primary segmentation goes out the hydrology
Ruler image.
3. a kind of hydrology ruler water level based on image analysis according to claim 2 identifies evaluation method, which is characterized in that
The step 1-1-1 comprising the following specific steps
Step 1-1-1-1: the color space of original image is converted by rgb space to HSV space, extract color, saturation degree,
Three kinds of components of brightness；
Step 1-1-1-2: global threshold is determined using Otsu algorithm for color component；
Step 1-1-1-3: according to the global threshold determined in step 1-1-1-2, binary conversion treatment is carried out to color component, tentatively
Show hydrology ruler；
Step 1-1-1-4: the bianry image obtained using the method processing step 1-1-1-3 of median filtering removes isolated noise；
The step 1-1-2 comprising the following specific steps
Step 1-1-2-1: being handled the obtained image of step 1-1-1-4 using the closed operation in morphological method, is restored
The connection region of hydrology ruler；
Step 1-1-2-2: opening operation is carried out to filter out thin in image to the image that step 1-1-2-1 is obtained；
Step 1-1-2-3: connected domain mark is carried out to the image that step 1-1-2-2 is obtained, is removed according to connection area size small
Connected domain leaves hydrology ruler connected domain；
The preliminary hydrology ruler segmentation of the step 1-1-3 the following steps are included:
Step 1-1-3-1: the hydrology ruler connected domain that step 1-1-2-3 is obtained is done into boundary rectangle；
Step 1-1-3-2: the bianry image and original image that step 1-1-3-1 is obtained carry out and operation；
Step 1-1-3-3；The hydrology ruler image that step 1-1-3-2 is obtained is cut, background is removed.
4. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, which is characterized in that
In the step 1, the adaptive component map that constructs realizes that the accurate positioning of hydrology ruler comprises the following specific steps that:
Step 1-2-1: defining component map according to the different colours of hydrology ruler, if hydrology ruler is red, defines component map are as follows:
C=1-e-p
If hydrology ruler is blue, component map is defined are as follows:
C=1-e-p
Wherein, C is component map, and R, G, B is hydrology ruler color component, and ω is the parameter for adjusting color weight, and p is component factor, e
For the truth of a matter of natural logrithm；
Step 1-2-2: calculating degree of isolation to hydrology ruler image using Otsu algorithm, comes using degree of isolation as standard adaptive true
Set the tone whole color weight parameter ω, and ω when degree of isolation maximum is as color weight；
Step 1-2-3: hydrology ruler inverted image is eliminated using the method for median filtering, realizes the accurate positioning of hydrology ruler.
5. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, which is characterized in that
In the step 1, further includes hydrology ruler Slant Rectify step, comprises the following specific steps that:
Step 1-3-1: edge detection is carried out to image obtained by step 1-2-3 using the sobel operator of vertical structure；
Step 1-3-2: using the straight line in Hough transform detection image, hydrology ruler vertical edge is detected；
Step 1-3-3: it is process object with hydrology ruler vertical edge, tilt angle is calculated, then according to tilt angle to the hydrology
Ruler image carries out Slant Rectify.
6. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, which is characterized in that
In the step 2, the artificial character and number divided on hydrology ruler, and by data enhancing technology EDS extended data set, then with
Machine is divided into training set and test set, comprises the following specific steps that:
Step 2-1-1: the character on artificial segmentation hydrology ruler image, including number, character " E " are labeled simultaneously, make sample
This library；
Step 2-1-2: the picture of segmentation is disturbed using change of scale, random cropping, horizontal vertical overturning, translation transformation, noise
Dynamic, affine transformation is handled, and last size is normalized to 40 × 40, as data set；
Step 2-1-3: the data set of acquisition is selected 1/5 as test set at random, remaining is as training set.
7. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, which is characterized in that
In the step 2, constructs convolutional neural networks and is trained and tests and comprise the following specific steps that:
Step 2-2-1: 7 layers of convolutional neural networks of building is followed successively by input layer, C1 convolutional layer, S2 down-sampling layer, C3 later in the past
Convolutional layer, S4 down-sampling layer, full articulamentum and output layer；
Step 2-2-2: input training dataset, and initialize network and start to train；
Step 2-2-3: input test data set tests trained network, and recording accuracy；
Wherein, the step 2-2-1 is comprised the following specific steps that:
Step 2-2-1-1: output characteristic pattern OutFeatureMaps, convolution kernel kernel and the biasing of each convolutional layer are determined
Bias, output characteristic pattern OutFeatureMaps includes characteristic pattern quantity OutFeatureMaps_Num and characteristic pattern size
OutFeatureMaps_Size；Convolution kernel kernel includes convolution kernel size Kernel_Size and convolution nuclear volume
Kernel_Num；The output characteristic pattern quantity OutFeatureMaps_Num phase of the quantity Bias_Num of offset parameter and respective layer
Together；
For convolutional layer Cx, x ∈ { 1,3 }, the output characteristic pattern width of respective layer is OutFeatureMaps_Size_Cx, value
Convolution kernel size Kernel_Size_Cx by its input feature vector figure resolution ratio InFeatureMaps_Size and respective layer is common
It determines, i.e. OutFeatureMaps_Size_Cx=InFeatureMaps_Size-Kernel_Size_Cx+1；
For convolutional layer C1, enabling it export characteristic pattern quantity is OutFeatureMaps_Num_C1=6, exports characteristic pattern width
For OutFeatureMaps_Size_C1=36；The convolution kernel size Kernel_Size_C1=5 of convolutional layer C1, convolutional layer C1's
Convolution nuclear volume Kernel_Num_C1=6 initializes convolution kernel numerical value using Xavier method, and initial value isBetween 5 rank matrixes；The quantity Bias_Num_C1=6 of the offset parameter of convolutional layer C1, initial value
It is 0；
For convolutional layer C3, enabling it export characteristic pattern quantity is OutFeatureMaps_Num_C1=12, exports characteristic pattern width
For OutFeatureMaps_Size_C1=14；The convolution kernel size Kernel_Size_C1=5 of convolutional layer C3, convolutional layer C3's
Convolution nuclear volume Kernel_Num_C1=12 initializes convolution kernel numerical value using Xavier method, and initial value isBetween 5 rank matrixes；The quantity Bias_Num_C1=12 of the offset parameter of convolutional layer C3, just
Initial value is 0；
Step 2-2-1-2: down-sampling layer be arranged sample size scale=2, using average pond method, i.e., by down-sampling layer S2,
The sampling core of S4 is initialized asThe output characteristic pattern quantity OutFeatureMaps_Num of sample level S2, S4 with
Its upper layer convolutional layer output characteristic pattern quantity is equal, and exporting characteristic pattern resolution ratio OutFeatureMaps_Size is its upper layer volume
Lamination exports characteristic pattern resolution ratio half, it may be assumed that
OutFeatureMaps_Num_S2=6, OutFeatureMaps_Size_S2=18,
OutFeatureMaps_Num_S4=12, OutFeatureMaps_Size_S2=7；
Step 2-2-1-3: constructing full articulamentum, and full articulamentum is connected with output layer, and weight parameter FW initial value isBetween size be 11 × 588 matrix, the size that offset parameter FB initial value is 0 is 11 × 1
Vector, each data finally export the probability value of all kinds of labels, take label corresponding to maximum probability for output result；
Wherein, the step 2-2-2 is comprised the following specific steps that:
Step 2-2-2-1: concentrating each secondary gray scale diagram data to be normalized between [0,1] by [0,255] training data, and with
Its corresponding label, which matches, is input to the input layer of network；
Step 2-2-2-2: the learning rate alpha=0.1 of setting network, setting batch size batchsize=5 set training time
Number iteration=1000；
Step 2-2-2-3: using sigmoid function as activation primitive, started using secondary cost function as loss function
Network training reaches frequency of training iteration deconditioning；
Wherein, the step 2-2-3 is comprised the following specific steps that:
Step 2-2-3-1: it concentrates each secondary gray scale diagram data to be normalized between [0,1] by [0,255] test data, inputs
It is tested to trained network；
Step 2-2-2-2: after starting test, outputing test result, and is compared with label corresponding to test set, and record is accurate
Degree.
8. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, which is characterized in that
In the step 2, hydrology ruler image is clustered and divided using the method for fuzzy C-means clustering, the character being partitioned into is used
Trained convolutional neural networks are identified, are comprised the following specific steps that:
Step 2-3-1: and hydrology ruler image is clustered using the method for fuzzy C-means clustering, by white background and number, character
It is clustered, to distinguish background and character；
Step 2-3-2: the character for completing cluster is split, after the completion of segmentation, according to sequence from top to bottom, from left to right
Arrangement, it is endless if last character is defined as imperfect character since water level blocks appearance incompleteness by last character
Whole character is incomplete number or incomplete character " E ", remaining is complete character, complete character include complete number and
Complete character " E "；
Step 2-3-3: the complete character that will be partitioned into, including complete number and complete character " E " normalize its size
To 40 × 40, identified using convolutional neural networks, integer part of the recognition result as the depth of water.
9. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, which is characterized in that
The step 3 comprises the following specific steps that:
Step 3-1: in order by the complete character identified in step 2 " E ", i.e., according to from top to bottom, from left to right suitable
Sequence storage, amounts to m character " E ", calculates its pixels tall, be denoted as hE1, hE2, hE3..., hEm；The last one incomplete character is made
It is m+1, pixels tall hE(m+1)；
Step 3-2: using the mapping relations between quadratic equation match pixel and actual height, by hE1, (hE1+hE2), (hE1+hE2
+hE3) ..., (hE1+hE2+hE3+…+hEm) it is used as independent variable x, represent accumulated pixel height；It corresponds to 50,100,150 ...,
50n represents actual height as dependent variable y, and unit is millimeter mm；To quadratic equation, i.e. y=ax2+ bx+c is fitted, and is asked
Solve coefficient a, b, c；
Step 3-3: the actual height that last incomplete character represents is calculated, x=(h is enabledE1+hE2+hE3+…+hEm+hE(m+1)), band
Enter equation y=ax2Y is solved in+bx+c, calculating last its actual height of incomplete character is (y-50m) mm.
10. a kind of hydrology ruler water level based on image analysis according to claim 1 identifies evaluation method, feature exists
In, in the step 4, be divided into integer part calculate and fractional part calculate；
Integer portion calculates: according to the positional relationship of water level line and hydrology ruler, with the left-half of hydrology ruler, i.e., comprising number and word
The half for according with " E " is standard, establishes two kinds of situations:
Situation 1: since water level line blocks, water level wire cutting character " E ", that is, forming hydrology ruler left-half last character is
Incomplete character " E ", then the last one complete character is number, recognition result n, primarily determine depth of water range be (10n-5,
10n) cm, and in this, as integer part；
Situation 2: since water level line blocks, water level wire cutting number, that is, it is residual for forming hydrology ruler right half part last character
Scarce number, then the last one complete character is character " E ", and penultimate character is digital, recognition result n, tentatively
Determine that depth of water range is (10n, 10n+5) cm, and in this, as integer part；
Fractional part calculates: if calculating the actual height of the last one incomplete character " E " by step 3 is h cm, according to integer
Portion's calculated result arranges the integer part and fractional part of the depth of water, calculates depth of water height, carries out depth of water calculating in two kinds of situation:
Situation 1: determine integer part depth of water range be (10n-5,10n) cm, then calculate hydrology ruler left-half the last one
The pixels tall of incomplete character " E ", and its actual height h cm is calculated using dynamic mapping, then last measurement result is (10n-
h)cm；
Situation 2: determine integer part depth of water range be (10n, 10n+5) cm, then calculate hydrology ruler right half part the last one
The pixels tall of incomplete character " E ", and its actual height h cm is calculated using dynamic mapping, then last measurement result is (10n+
5-h)cm。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811020954.3A CN109522889A (en) | 2018-09-03 | 2018-09-03 | Hydrological ruler water level identification and estimation method based on image analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811020954.3A CN109522889A (en) | 2018-09-03 | 2018-09-03 | Hydrological ruler water level identification and estimation method based on image analysis |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109522889A true CN109522889A (en) | 2019-03-26 |
Family
ID=65770753
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811020954.3A Pending CN109522889A (en) | 2018-09-03 | 2018-09-03 | Hydrological ruler water level identification and estimation method based on image analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109522889A (en) |
Cited By (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110427933A (en) * | 2019-06-20 | 2019-11-08 | 浙江大学 | A kind of water gauge recognition methods based on deep learning |
| CN110443243A (en) * | 2019-08-07 | 2019-11-12 | 浙江大华技术股份有限公司 | Water level monitoring method, storage medium, the network equipment and water level monitoring system |
| CN110472636A (en) * | 2019-07-26 | 2019-11-19 | 四创科技有限公司 | Water gauge E font scale recognition methods based on deep learning |
| CN111027546A (en) * | 2019-12-05 | 2020-04-17 | 北京嘉楠捷思信息技术有限公司 | Character segmentation method and device and computer readable storage medium |
| CN111259890A (en) * | 2020-01-19 | 2020-06-09 | 深圳市宏电技术股份有限公司 | Water level identification method, device and equipment of water level gauge |
| CN111626190A (en) * | 2020-05-26 | 2020-09-04 | 浙江大学 | Water level monitoring method for scale recognition based on clustering partitions |
| CN111652089A (en) * | 2020-05-18 | 2020-09-11 | 纪义胜 | Water level automatic identification method and system based on image processing |
| CN111950530A (en) * | 2020-09-08 | 2020-11-17 | 中国水利水电科学研究院 | Multi-feature optimization and fusion method for extracting crop planting structure |
| CN112013921A (en) * | 2019-05-30 | 2020-12-01 | 杭州海康威视数字技术股份有限公司 | Method, device and system for acquiring water level information based on water level gauge measurement image |
| WO2021109697A1 (en) * | 2019-12-05 | 2021-06-10 | 嘉楠明芯(北京)科技有限公司 | Character segmentation method and apparatus, and computer-readable storage medium |
| CN113823088A (en) * | 2021-09-11 | 2021-12-21 | 淮阴工学院 | Urban road ponding depth prediction and early warning method based on visual recognition technology |
| CN114663811A (en) * | 2022-03-24 | 2022-06-24 | 中国水利水电科学研究院 | Water surface line extraction method using water gauge reflection |
| CN115546793A (en) * | 2022-12-05 | 2022-12-30 | 安徽大学 | Water gauge scale automatic reading method and system and electronic equipment |
| CN115578695A (en) * | 2022-11-21 | 2023-01-06 | 昆明理工大学 | Water gauge water level machine vision detection method and device with free shooting visual angle |
| CN116342965A (en) * | 2023-05-26 | 2023-06-27 | 中国电建集团江西省电力设计院有限公司 | Water level measurement error analysis and control method and system |
| CN116385735A (en) * | 2023-06-01 | 2023-07-04 | 珠江水利委员会珠江水利科学研究院 | Water level measurement method based on image recognition |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH11304568A (en) * | 1998-04-17 | 1999-11-05 | Hitachi Ltd | Liquid level measuring apparatus |
| CN102928051A (en) * | 2012-11-01 | 2013-02-13 | 清华大学 | Water gauge suitable for video detection of water level |
| CN102494733B (en) * | 2011-12-12 | 2013-09-25 | 西安电子科技大学 | Water level monitoring system based on image processing and method |
| CN103886612A (en) * | 2014-04-08 | 2014-06-25 | 马鸿旭 | Automatic water level extraction method and system based on reservoir monitoring camera |
| CN106557764A (en) * | 2016-11-02 | 2017-04-05 | 江西理工大学 | A kind of water level recognition methodss based on binary-coded character water gauge and image procossing |
| CN107367310A (en) * | 2017-07-11 | 2017-11-21 | 华南理工大学 | A kind of river level remote monitoring method based on computer vision |
| CN108073905A (en) * | 2017-12-21 | 2018-05-25 | 北京邮电大学 | A kind of method, system and the equipment of intelligence water gauge reading |
-
2018
- 2018-09-03 CN CN201811020954.3A patent/CN109522889A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH11304568A (en) * | 1998-04-17 | 1999-11-05 | Hitachi Ltd | Liquid level measuring apparatus |
| CN102494733B (en) * | 2011-12-12 | 2013-09-25 | 西安电子科技大学 | Water level monitoring system based on image processing and method |
| CN102928051A (en) * | 2012-11-01 | 2013-02-13 | 清华大学 | Water gauge suitable for video detection of water level |
| CN103886612A (en) * | 2014-04-08 | 2014-06-25 | 马鸿旭 | Automatic water level extraction method and system based on reservoir monitoring camera |
| CN106557764A (en) * | 2016-11-02 | 2017-04-05 | 江西理工大学 | A kind of water level recognition methodss based on binary-coded character water gauge and image procossing |
| CN107367310A (en) * | 2017-07-11 | 2017-11-21 | 华南理工大学 | A kind of river level remote monitoring method based on computer vision |
| CN108073905A (en) * | 2017-12-21 | 2018-05-25 | 北京邮电大学 | A kind of method, system and the equipment of intelligence water gauge reading |
Non-Patent Citations (6)
| Title |
|---|
| MACHIEL BRUININK ET AL: "Portable, automatic water level estimation using mobile phone cameras", 《14TH IAPR INTERNATIONAL CONFERENCE ON MACHINE VISION APPLICATIONS (MVA)》 * |
| 姜兵: "基于图像识别的远程水位监测系统研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 徐志康等: "道路积水深度测量研究综述", 《电子测量技术》 * |
| 李德毅等: "《中国科协新一代信息技术系列丛书 人工智能导论》", 30 August 2018 * |
| 王欣蔚: "基于深度学习的TEDS系统车号识别算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
| 高志强等: "《深度学习从入门到实践》", 30 June 2018 * |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112013921A (en) * | 2019-05-30 | 2020-12-01 | 杭州海康威视数字技术股份有限公司 | Method, device and system for acquiring water level information based on water level gauge measurement image |
| CN110427933A (en) * | 2019-06-20 | 2019-11-08 | 浙江大学 | A kind of water gauge recognition methods based on deep learning |
| CN110472636A (en) * | 2019-07-26 | 2019-11-19 | 四创科技有限公司 | Water gauge E font scale recognition methods based on deep learning |
| CN110472636B (en) * | 2019-07-26 | 2022-10-14 | 四创科技有限公司 | Deep learning-based water gauge E-shaped scale identification method |
| CN110443243A (en) * | 2019-08-07 | 2019-11-12 | 浙江大华技术股份有限公司 | Water level monitoring method, storage medium, the network equipment and water level monitoring system |
| CN110443243B (en) * | 2019-08-07 | 2022-06-07 | 浙江大华技术股份有限公司 | Water level monitoring method, storage medium, network device and water level monitoring system |
| WO2021109697A1 (en) * | 2019-12-05 | 2021-06-10 | 嘉楠明芯(北京)科技有限公司 | Character segmentation method and apparatus, and computer-readable storage medium |
| CN111027546A (en) * | 2019-12-05 | 2020-04-17 | 北京嘉楠捷思信息技术有限公司 | Character segmentation method and device and computer readable storage medium |
| CN111027546B (en) * | 2019-12-05 | 2024-03-26 | 嘉楠明芯(北京)科技有限公司 | Character segmentation method, device and computer readable storage medium |
| CN111259890A (en) * | 2020-01-19 | 2020-06-09 | 深圳市宏电技术股份有限公司 | Water level identification method, device and equipment of water level gauge |
| CN111652089A (en) * | 2020-05-18 | 2020-09-11 | 纪义胜 | Water level automatic identification method and system based on image processing |
| CN111652089B (en) * | 2020-05-18 | 2023-09-19 | 纪义胜 | Automatic water level identification method and system based on image processing |
| CN111626190A (en) * | 2020-05-26 | 2020-09-04 | 浙江大学 | Water level monitoring method for scale recognition based on clustering partitions |
| CN111950530A (en) * | 2020-09-08 | 2020-11-17 | 中国水利水电科学研究院 | Multi-feature optimization and fusion method for extracting crop planting structure |
| CN111950530B (en) * | 2020-09-08 | 2024-04-12 | 中国水利水电科学研究院 | Multi-feature optimization and fusion method for crop planting structure extraction |
| CN113823088A (en) * | 2021-09-11 | 2021-12-21 | 淮阴工学院 | Urban road ponding depth prediction and early warning method based on visual recognition technology |
| CN114663811A (en) * | 2022-03-24 | 2022-06-24 | 中国水利水电科学研究院 | Water surface line extraction method using water gauge reflection |
| CN115578695A (en) * | 2022-11-21 | 2023-01-06 | 昆明理工大学 | Water gauge water level machine vision detection method and device with free shooting visual angle |
| CN115546793A (en) * | 2022-12-05 | 2022-12-30 | 安徽大学 | Water gauge scale automatic reading method and system and electronic equipment |
| CN116342965A (en) * | 2023-05-26 | 2023-06-27 | 中国电建集团江西省电力设计院有限公司 | Water level measurement error analysis and control method and system |
| CN116342965B (en) * | 2023-05-26 | 2023-11-24 | 中国电建集团江西省电力设计院有限公司 | Water level measurement error analysis and control method and system |
| CN116385735A (en) * | 2023-06-01 | 2023-07-04 | 珠江水利委员会珠江水利科学研究院 | Water level measurement method based on image recognition |
| CN116385735B (en) * | 2023-06-01 | 2023-09-01 | 珠江水利委员会珠江水利科学研究院 | Water level measurement method based on image recognition |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109522889A (en) | Hydrological ruler water level identification and estimation method based on image analysis | |
| CN109145830B (en) | Intelligent water gauge identification method | |
| Huang et al. | Morphological building/shadow index for building extraction from high-resolution imagery over urban areas | |
| Ke et al. | A review of methods for automatic individual tree-crown detection and delineation from passive remote sensing | |
| CN103035013B (en) | A kind of precise motion shadow detection method based on multi-feature fusion | |
| Heinzel et al. | Prior-knowledge-based single-tree extraction | |
| CN110119728A (en) | Remote sensing images cloud detection method of optic based on Multiscale Fusion semantic segmentation network | |
| CN109344701A (en) | A kind of dynamic gesture identification method based on Kinect | |
| Gao et al. | Building extraction from RGB VHR images using shifted shadow algorithm | |
| CN109886238A (en) | Unmanned plane Image Change Detection algorithm based on semantic segmentation | |
| Qu et al. | A pedestrian detection method based on yolov3 model and image enhanced by retinex | |
| CN104077577A (en) | Trademark detection method based on convolutional neural network | |
| CN108960404B (en) | Image-based crowd counting method and device | |
| CN108257151B (en) | PCANet image change detection method based on significance analysis | |
| CN108388905B (en) | A kind of Illuminant estimation method based on convolutional neural networks and neighbourhood context | |
| CN105389581B (en) | A kind of rice germ plumule integrity degree intelligent identifying system and its recognition methods | |
| CN106874929B (en) | Pearl classification method based on deep learning | |
| CN110263768A (en) | A kind of face identification method based on depth residual error network | |
| CN106934386A (en) | A kind of natural scene character detecting method and system based on from heuristic strategies | |
| CN107038416A (en) | A kind of pedestrian detection method based on bianry image modified HOG features | |
| CN109635799B (en) | Method for recognizing number of character wheel of gas meter | |
| Wang et al. | Color-and texture-based image segmentation for improved forest delineation | |
| Lambers et al. | Towards detection of archaeological objects in high-resolution remotely sensed images: the Silvretta case study | |
| CN111291818B (en) | Non-uniform class sample equalization method for cloud mask | |
| CN108280483A (en) | Trypetid adult image-recognizing method based on neural network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190326 |