CN115294486A - Method for identifying violation building data based on unmanned aerial vehicle and artificial intelligence - Google Patents

Method for identifying violation building data based on unmanned aerial vehicle and artificial intelligence Download PDF

Info

Publication number
CN115294486A
CN115294486A CN202211219087.2A CN202211219087A CN115294486A CN 115294486 A CN115294486 A CN 115294486A CN 202211219087 A CN202211219087 A CN 202211219087A CN 115294486 A CN115294486 A CN 115294486A
Authority
CN
China
Prior art keywords
artificial intelligence
unmanned aerial
aerial vehicle
image
value
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.)
Granted
Application number
CN202211219087.2A
Other languages
Chinese (zh)
Other versions
CN115294486B (en
Inventor
陈钢
刘攀
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Byte Technology Qingdao Co Ltd
Original Assignee
Byte Technology Qingdao Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Byte Technology Qingdao Co Ltd filed Critical Byte Technology Qingdao Co Ltd
Priority to CN202211219087.2A priority Critical patent/CN115294486B/en
Publication of CN115294486A publication Critical patent/CN115294486A/en
Application granted granted Critical
Publication of CN115294486B publication Critical patent/CN115294486B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/17—Terrestrial scenes taken from planes or by drones
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • Artificial Intelligence (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Remote Sensing (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to the technical field of image processing, in particular to a violation building data identification method based on an unmanned aerial vehicle and artificial intelligence, which comprises the following steps: collecting a high-definition video by an unmanned aerial vehicle; performing interframe difference processing on a high-definition video acquired by an unmanned aerial vehicle to acquire an effective high-definition picture; transmitting the obtained high-definition picture to a ground workstation by a 5G technology, and further preprocessing the picture; and (4) carrying out operation on the preprocessed pictures through an artificial intelligence algorithm, and further realizing the identification and judgment of the garbage. According to the invention, the unmanned aerial vehicle is used for acquiring image data, and by improving the frame difference algorithm and expanding the target with slow frame difference detection change, useless or repeated pictures discarded by effective high-definition pictures can be obtained, and the 5G transmission speed and the detection efficiency of the artificial intelligence algorithm are further improved; the garbage data image is combined with an artificial intelligence algorithm to realize automatic recognition and judgment of garbage.

Description

Method for identifying violation building data based on unmanned aerial vehicle and artificial intelligence
Technical Field
The invention relates to the technical field of image processing, in particular to a violation building data identification method based on an unmanned aerial vehicle and artificial intelligence.
Background
In the past, informatization technology is not common, and the illegal building management department usually adopts a manual inspection means to manually find rubbish. In recent years, a method for performing remote monitoring by using a camera is also presented for monitoring nearby illegal buildings, but the method has some defects, such as dead monitoring corners, large capital investment and the like. Although the methods can be used for collecting the garbage images, the garbage identification and judgment are mainly carried out in a manual identification mode, when the number of the obtained images is large or the image range is large, huge workload can be generated in the manual identification mode, and meanwhile, the identification efficiency is relatively low. And the current technologies such as intelligent identification and automatic feature extraction are still in the research stage and cannot be widely applied, so that the unmanned aerial vehicle inspection effect is greatly reduced.
Disclosure of Invention
The invention aims to solve the defects in the background technology by providing a violation building data identification method based on an unmanned aerial vehicle and artificial intelligence.
The technical scheme adopted by the invention is as follows:
the method for identifying the violation building data based on the unmanned aerial vehicle and the artificial intelligence comprises the following steps:
s1.1: collecting a high-definition video by an unmanned aerial vehicle;
s1.2: performing interframe difference processing on a high-definition video acquired by an unmanned aerial vehicle to acquire an effective high-definition picture;
s1.3: transmitting the obtained high-definition picture to a ground workstation by a 5G technology, and further preprocessing the picture;
s1.4: and (4) calculating the preprocessed pictures through an artificial intelligence algorithm, and further realizing the identification and judgment of the garbage.
As a preferred technical scheme of the invention: in the step S1.2, an improved inter-frame difference method is adopted for inter-frame difference processing, and the improved inter-frame difference method performs difference operation on images of a current frame and previous and next frames by using three frames of image information.
As a preferred technical scheme of the invention: the improved interframe difference method formula is as follows:
Figure 184048DEST_PATH_IMAGE001
；
Figure 288139DEST_PATH_IMAGE002
and
Figure 5559DEST_PATH_IMAGE003
the images are respectively the difference between the front frame and the back frame and the current frame,
Figure 400769DEST_PATH_IMAGE004
and
Figure 402091DEST_PATH_IMAGE005
、
Figure 162237DEST_PATH_IMAGE006
the gray values of the image at time t, time t-1 and time t +1 respectively,
Figure 616221DEST_PATH_IMAGE007
is a coefficient of gray scale to be used,
Figure 815121DEST_PATH_IMAGE008
is the total number of pixels of the region to be detected,
Figure 421683DEST_PATH_IMAGE009
an image to be detected is obtained;
and then the frame difference image is obtained through threshold value T processing, and binarization processing is carried out:
Figure 601997DEST_PATH_IMAGE010
；
wherein,
Figure 356327DEST_PATH_IMAGE011
or
Figure 31022DEST_PATH_IMAGE012
，
Figure 741358DEST_PATH_IMAGE013
Or all are binarized toThe results obtained.
As a preferred technical scheme of the invention: in the S1.3, the preprocessing method comprises image gray processing, color inversion and a canny edge detection algorithm;
the picture gray level processing formula is as follows:
Figure 108885DEST_PATH_IMAGE014
；
wherein,
Figure 350511DEST_PATH_IMAGE015
、
Figure 812585DEST_PATH_IMAGE016
and
Figure 393739DEST_PATH_IMAGE017
respectively representing components of red, green and blue, and averaging the three components of the color image to obtain a gray value;
the color inversion formula is as follows:
Figure 915856DEST_PATH_IMAGE018
；
wherein,
Figure 644778DEST_PATH_IMAGE019
for the value of the current pixel, it is,
Figure 661275DEST_PATH_IMAGE020
is the pixel value after color inversion;
the canny edge detection algorithm is as follows:
Figure 346203DEST_PATH_IMAGE021
；
Figure 117850DEST_PATH_IMAGE022
is a Gaussian distributionThe standard deviation of the measured data was found to be,
Figure 6172DEST_PATH_IMAGE023
is a pixel point;
multiplying each pixel point and the neighborhood thereof by a Gaussian matrix, and taking the weighted average value as the final gray value:
Figure 75628DEST_PATH_IMAGE024
；
Figure 428112DEST_PATH_IMAGE025
；
wherein,
Figure 308343DEST_PATH_IMAGE026
is the amplitude of the received signal and is,
Figure 933228DEST_PATH_IMAGE027
in the form of a direction of rotation,
Figure 291528DEST_PATH_IMAGE028
and
Figure 764098DEST_PATH_IMAGE029
respectively the image at a pixel point
Figure 64498DEST_PATH_IMAGE030
Horizontal gradient magnitude and vertical gradient magnitude.
As a preferred technical scheme of the invention: after the canny edge detection algorithm is detected, calculating a gradient value and a gradient direction through a sobel operator, and filtering out a maximum value; setting eight gradient directions, including:
Figure 927412DEST_PATH_IMAGE031
；
and setting a threshold value:
Figure 338671DEST_PATH_IMAGE032
；
wherein,
Figure 400168DEST_PATH_IMAGE033
gray values of all points to be measured;
and filtering the noise according to the image definition evaluation value:
Figure 622202DEST_PATH_IMAGE034
；
wherein,
Figure 221679DEST_PATH_IMAGE035
is effective.
As a preferred technical scheme of the invention: in S1.4, the artificial intelligence algorithm includes: image input, basic feature extraction, multi-layer complex feature extraction, feature learning and classification detection results; the method comprises the steps of carrying out convolution calculation on an image, outputting a classification result by a Softmax classifier through a pooling layer, an activation function and a full connection layer, and realizing the identification and judgment of garbage.
As a preferred technical scheme of the invention: the convolution calculation is that a small matrix slides on an image or an input characteristic diagram, the output characteristic diagram is a result obtained by multiplying and adding the matrix, and the calculation method comprises the following steps:
Figure 249678DEST_PATH_IMAGE036
；
wherein,
Figure 368944DEST_PATH_IMAGE037
is the number of convolution kernel channels, sum is the matrix addition operator,bin order to be a characteristic parameter of the device,
performing aggregation statistics through the features of different positions, and selecting a representative value to represent the original feature; by adopting the maxpololing method, the calculation formula is as follows:
Figure 745567DEST_PATH_IMAGE038
；
wherein,
Figure 583073DEST_PATH_IMAGE039
characteristic values of different positions.
Introducing an activation function relu function in the neural network, wherein the calculation formula is as follows:
Figure 149184DEST_PATH_IMAGE040
；
wherein x is a characteristic value used for improving the characterization capability of the artificial intelligence algorithm.
As a preferred technical scheme of the invention: the convolution layer, the pooling layer and the activation function structure map original data to a feature vector space, and the full connection layer is obtained by a calculation formula:
Figure 637803DEST_PATH_IMAGE041
；
wherein,
Figure 936060DEST_PATH_IMAGE042
for the mapped feature samples, the learned distributed features are integrated and summarized and then mapped to a sample label space.
As a preferred technical scheme of the invention: softmax maps the feature vectors of the input neural network to (0, 1) space, the sum of these values is 1, and the maximum probability value is selected as the classification result:
Figure 323179DEST_PATH_IMAGE043
；
wherein,
Figure 879931DEST_PATH_IMAGE044
is a classification vector.
As a preferred technical scheme of the invention: and after the artificial intelligence algorithm is processed, performing garbage detection, and summarizing detection results to terminal equipment for visual display and information storage.
Compared with the prior art, the violation building data identification method based on the unmanned aerial vehicle and the artificial intelligence has the beneficial effects that:
according to the invention, the unmanned aerial vehicle is used for acquiring image data, and by improving the frame difference algorithm and expanding the target with slow frame difference detection change, useless or repeated pictures discarded by effective high-definition pictures can be obtained, and the 5G transmission speed and the detection efficiency of the artificial intelligence algorithm are further improved; the garbage data image is combined with an artificial intelligence algorithm to realize automatic recognition and judgment of garbage. The unmanned aerial photography technology is used for photography, and a remote sensing platform which is convenient to operate and easy to transfer is provided for aerial photography. The take-off and landing are less limited by the field, and the landing can be carried out on playgrounds, highways or other open grounds, so that the take-off and landing are good in stability and safety, and the take-off and landing are very easy to transfer.
Drawings
FIG. 1 is a flow chart of a method of a preferred embodiment of the present invention;
fig. 2 is a technical structural view of a preferred embodiment of the present invention.
Detailed Description
It should be noted that, in the case of no conflict, the embodiments and features in the embodiments may be combined with each other, and the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Referring to fig. 1, a preferred embodiment of the present invention provides a violation building data identification method based on unmanned aerial vehicles and artificial intelligence, including the following steps:
s1.1: collecting a high-definition video by an unmanned aerial vehicle;
s1.2: performing interframe difference processing on the high-definition video acquired by the unmanned aerial vehicle to acquire an effective high-definition picture;
s1.3: transmitting the obtained high-definition picture to a ground workstation by a 5G technology, and further preprocessing the picture;
s1.4: and (4) carrying out operation on the preprocessed pictures through an artificial intelligence algorithm, and further realizing the identification and judgment of the garbage.
In the step S1.2, an improved inter-frame difference method is adopted for inter-frame difference processing, and the improved inter-frame difference method performs difference operation on images of a current frame and previous and next frames by using three frames of image information.
The improved interframe difference method formula is as follows:
Figure 708210DEST_PATH_IMAGE045
；
Figure 692215DEST_PATH_IMAGE046
and
Figure 301051DEST_PATH_IMAGE047
the images are respectively the difference between the front frame and the back frame and the current frame,
Figure 412227DEST_PATH_IMAGE004
and
Figure 609859DEST_PATH_IMAGE005
、
Figure 577815DEST_PATH_IMAGE006
the gray values of the image at the time t, the time t-1 and the time t +1 respectively,
Figure 611630DEST_PATH_IMAGE007
is a function of the gamma of the color,
Figure 510185DEST_PATH_IMAGE008
is the total number of pixels of the region to be detected,
Figure 313055DEST_PATH_IMAGE009
for the image to be detected；
And then the frame difference image is obtained through threshold value T processing, and binarization processing is carried out:
Figure 451913DEST_PATH_IMAGE048
；
wherein,
Figure 222292DEST_PATH_IMAGE011
or
Figure 409690DEST_PATH_IMAGE012
，
Figure 394964DEST_PATH_IMAGE013
Or both results are obtained by binarization.
In the S1.3, the preprocessing method comprises image gray processing, color inversion and a canny edge detection algorithm;
the picture gray level processing formula is as follows:
Figure 891673DEST_PATH_IMAGE049
wherein,
Figure 900081DEST_PATH_IMAGE015
、
Figure 140438DEST_PATH_IMAGE016
and
Figure 980218DEST_PATH_IMAGE017
respectively representing components of red, green and blue, and averaging the three components of the color image to obtain a gray value;
the color inversion formula is as follows:
Figure 398561DEST_PATH_IMAGE050
；
wherein,
Figure 877953DEST_PATH_IMAGE019
for the value of the current pixel, it is,
Figure 672733DEST_PATH_IMAGE051
the pixel values after color inversion;
the canny edge detection algorithm is as follows:
Figure 367020DEST_PATH_IMAGE052
；
wherein,
Figure 205532DEST_PATH_IMAGE053
is the standard deviation of the Gaussian distribution,
Figure 188531DEST_PATH_IMAGE030
the pixel points are set;
multiplying each pixel point and the neighborhood thereof by a Gaussian matrix, and taking the weighted average value as the final gray value:
Figure 583740DEST_PATH_IMAGE054
；
Figure 319484DEST_PATH_IMAGE055
；
wherein,
Figure 345209DEST_PATH_IMAGE026
in order to be the amplitude value,
Figure 799193DEST_PATH_IMAGE027
in the form of a direction of rotation,
Figure 998093DEST_PATH_IMAGE028
and
Figure 339076DEST_PATH_IMAGE029
respectively the image at a pixel point
Figure 620907DEST_PATH_IMAGE030
Horizontal gradient magnitude and vertical gradient magnitude.
After the canny edge detection algorithm is detected, calculating a gradient value and a gradient direction through a sobel operator, and filtering out a maximum value; setting eight gradient directions, including:
Figure 312919DEST_PATH_IMAGE031
；
and setting a threshold value:
Figure 49931DEST_PATH_IMAGE056
；
wherein,
Figure 760267DEST_PATH_IMAGE057
gray values of all points to be measured;
and filtering the noise according to the image definition evaluation value:
Figure 127795DEST_PATH_IMAGE058
；
wherein,
Figure 290792DEST_PATH_IMAGE059
is effective.
In S1.4, the artificial intelligence algorithm includes: image input, basic feature extraction, multi-layer complex feature extraction, feature learning and classification detection results; the method comprises the steps of carrying out convolution calculation on an image, outputting a classification result by a Softmax classifier through a pooling layer, an activation function and a full connection layer, and realizing the identification and judgment of the garbage.
The convolution calculation is that a small matrix slides on an image or an input characteristic diagram, the output characteristic diagram is a result obtained by multiplication and addition of the matrix, and the calculation method is as follows:
Figure 769177DEST_PATH_IMAGE060
；
wherein,
Figure 412648DEST_PATH_IMAGE061
is the number of convolution kernel channels, sum is the matrix addition operator,bin order to be a characteristic parameter of the device,
performing aggregation statistics through the features of different positions, and selecting a representative value to represent the original feature; by adopting the maxporoling method, the calculation formula is as follows:
Figure 934765DEST_PATH_IMAGE062
；
wherein,
Figure 601370DEST_PATH_IMAGE063
characteristic values of different positions.
Introducing an activation function relu function in the neural network, wherein the calculation formula is as follows:
Figure 867135DEST_PATH_IMAGE064
；
wherein x is a characteristic value used for improving the characterization capability of the artificial intelligence algorithm.
The convolution layer, the pooling layer and the activation function structure map original data to a feature vector space, and the full connection layer is obtained by a calculation formula:
Figure 365113DEST_PATH_IMAGE065
；
wherein,
Figure 808864DEST_PATH_IMAGE066
for the mapped feature samples, the learned distributed features are summarized in an integration and then mapped to a sample label space.
Softmax maps the feature vectors of the input neural network to (0, 1) space, and the sum of these values is 1, selecting the maximum probability value as the classification result:
Figure 212032DEST_PATH_IMAGE067
；
wherein,
Figure 32220DEST_PATH_IMAGE068
is a classification vector.
And after the artificial intelligence algorithm is processed, performing garbage detection, and summarizing detection results to terminal equipment for visual display and information storage.
In this embodiment, referring to fig. 2, the garbage data identification algorithm based on the unmanned aerial vehicle and the artificial intelligence mainly comprises three parts: unmanned aerial vehicle, 5G technology and artificial intelligence algorithm. The artificial intelligence algorithm comprises image input, basic feature extraction, multi-layer complex feature extraction, feature learning and classification detection results.
The unmanned aerial vehicle plays a role in collecting high-definition videos, interframe difference processing is carried out on the videos, effective high-definition pictures are obtained, useless or repeated pictures are abandoned, and the 5G transmission speed and the detection efficiency of an artificial intelligence algorithm are further improved. The implementation principle mathematical formula of the interframe difference method is expressed as follows:
Figure 384704DEST_PATH_IMAGE069
；
wherein,
Figure 514203DEST_PATH_IMAGE070
is a difference image between two successive frame images,
Figure 889821DEST_PATH_IMAGE071
and
Figure 310438DEST_PATH_IMAGE072
are respectively as
Figure 969958DEST_PATH_IMAGE073
And
Figure 755512DEST_PATH_IMAGE074
the image of the moment in time,
Figure 867693DEST_PATH_IMAGE075
is a threshold value selected when the difference image is binarized,
Figure 92001DEST_PATH_IMAGE076
the representation of the foreground is performed,
Figure 356760DEST_PATH_IMAGE077
representing the background.
The method is characterized in that an inter-frame difference method is improved, and a target with slow change is detected by expanding a frame difference, wherein the improved inter-frame difference method utilizes three frames of image information and carries out difference operation on images of a current frame and a front frame and a rear frame.
The improved interframe difference method formula is as follows:
Figure 828062DEST_PATH_IMAGE078
；
Figure 240588DEST_PATH_IMAGE046
and
Figure 940691DEST_PATH_IMAGE047
the images are respectively the difference between the front frame and the back frame and the current frame,
Figure 309225DEST_PATH_IMAGE004
and
Figure 702160DEST_PATH_IMAGE005
、
Figure 601983DEST_PATH_IMAGE006
the gray values of the image at the time t, the time t-1 and the time t +1 respectively,
Figure 355044DEST_PATH_IMAGE007
is a coefficient of gray scale to be used,
Figure 328816DEST_PATH_IMAGE008
is the total number of pixels of the region to be detected,
Figure 141920DEST_PATH_IMAGE009
an image to be detected is obtained;
and then, performing binarization processing on the frame difference image obtained by threshold T processing:
Figure 529039DEST_PATH_IMAGE079
；
wherein,
Figure 570945DEST_PATH_IMAGE011
or
Figure 914070DEST_PATH_IMAGE012
，
Figure 711125DEST_PATH_IMAGE013
Or both results are obtained by binarization.
And transmitting the image processed by the interframe difference method to a ground workstation by using a 5G technology, and further preprocessing the image, wherein the preprocessing method comprises image gray processing, color inversion and canny edge detection.
The mathematical expression formula of the picture gray processing is as follows:
Figure 257644DEST_PATH_IMAGE080
；
wherein,
Figure 618087DEST_PATH_IMAGE015
、
Figure 566451DEST_PATH_IMAGE016
and
Figure 534407DEST_PATH_IMAGE017
respectively represent components of red, green and blueThe three component luminance in the color image is averaged to obtain a gray value.
The color inversion mathematical expression is as follows:
Figure 817490DEST_PATH_IMAGE081
；
wherein,
Figure 201198DEST_PATH_IMAGE019
for the value of the current pixel, it is,
Figure 331965DEST_PATH_IMAGE051
the pixel values after color inversion; the inverted pixel value is equal to 255 minus the current pixel value.
The mathematical expression for canny edge detection is as follows:
Figure 657773DEST_PATH_IMAGE082
wherein,
Figure 178884DEST_PATH_IMAGE053
is the standard deviation of the Gaussian distribution,
Figure 615551DEST_PATH_IMAGE030
for pixel points, multiplying each pixel point and the neighborhood thereof by a Gaussian matrix, and taking the weighted average value as the final gray value:
Figure 600824DEST_PATH_IMAGE083
；
Figure 848266DEST_PATH_IMAGE084
；
wherein,
Figure 105941DEST_PATH_IMAGE026
in order to be the amplitude value,
Figure 97030DEST_PATH_IMAGE027
in the form of a direction of rotation,
Figure 936810DEST_PATH_IMAGE028
and
Figure 604421DEST_PATH_IMAGE029
respectively the image at a pixel point
Figure 834545DEST_PATH_IMAGE030
Horizontal gradient magnitude and vertical gradient magnitude.
Calculating a gradient value and a gradient direction through a sobel operator, and filtering out a maximum value; setting eight gradient directions, including:
Figure 691643DEST_PATH_IMAGE031
；
and setting a threshold value:
Figure 572880DEST_PATH_IMAGE085
；
wherein,
Figure 162124DEST_PATH_IMAGE086
gray values of all points to be measured;
and filtering the noise according to the image definition evaluation value:
Figure 394391DEST_PATH_IMAGE087
；
wherein,
Figure 789601DEST_PATH_IMAGE088
is effective.
And (3) sending the preprocessed image to an artificial intelligence algorithm, performing convolution calculation on the image, and outputting a classification result by a Softmax classifier after passing through a pooling layer, an activation function and a full connection layer. And then realize discernment and judgement to rubbish.
Convolution calculation is that some small matrixes slide on an image or an input characteristic diagram, the result obtained by multiplication and addition of the matrixes is an output characteristic diagram, and the calculation method comprises the following steps:
Figure 276077DEST_PATH_IMAGE089
；
wherein,
Figure 551069DEST_PATH_IMAGE061
is the number of convolution kernel channels, sum is the matrix addition operator,bis a characteristic parameter.
The feature size extracted after convolutional layer is still too large, and it is very inconvenient to directly use for training and easy to overfit. And performing aggregate statistics on the features at different positions, and selecting a representative value to represent the original feature. The model selection is a maxporoling method, and the calculation formula is as follows:
Figure 818102DEST_PATH_IMAGE090
；
wherein,
Figure 689106DEST_PATH_IMAGE091
characteristic values of different positions.
In order for an artificial intelligence algorithm to have good characterization capabilities, nonlinear elements must be introduced. Therefore, an activation function is introduced in the neural network. The activation function introduced by the model is a relu function, and the calculation formula is as follows:
Figure 544936DEST_PATH_IMAGE092
；
wherein x is a characteristic value.
The convolution layer, the pooling layer, the activation function and other structures map original data to a feature vector space, the full-connection layer is used for integrating and summarizing learned distributed features and then mapping the integrated and summarized distributed features to a sample mark space, and a calculation formula is as follows:
Figure 741562DEST_PATH_IMAGE093
；
wherein,
Figure 230312DEST_PATH_IMAGE094
are mapped feature samples.
Softmax maps the feature vectors of the input neural network to the (0, 1) space, and the sum of these values is 1, the output value can be understood as a probability value. So when outputting the result, the classification result with the highest probability value is selected.
Figure 419854DEST_PATH_IMAGE095
；
Wherein,
Figure 880922DEST_PATH_IMAGE044
is a classification vector.
After the processing of the artificial intelligence algorithm, the garbage can be accurately detected, and the detection result is gathered to the terminal equipment for visual display and information storage.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Furthermore, it should be understood that although the present description refers to embodiments, not every embodiment may contain only a single embodiment, and such description is for clarity only, and those skilled in the art should integrate the description, and the embodiments may be combined as appropriate to form other embodiments understood by those skilled in the art.

Claims (10)

1. The utility model provides a building data recognition method violating regulations based on unmanned aerial vehicle and artificial intelligence which characterized in that: the method comprises the following steps:
s1.1: collecting a high-definition video by an unmanned aerial vehicle;
s1.2: performing interframe difference processing on the high-definition video acquired by the unmanned aerial vehicle to acquire an effective high-definition picture;
s1.3: transmitting the obtained high-definition pictures to a ground workstation by a 5G technology, and further preprocessing the pictures;
s1.4: and (4) calculating the preprocessed pictures through an artificial intelligence algorithm, and further realizing the identification and judgment of the garbage.
2. The unmanned aerial vehicle and artificial intelligence based violation building data identification method as recited in claim 1, wherein: in the step S1.2, an improved inter-frame difference method is adopted for inter-frame difference processing, and the improved inter-frame difference method performs difference operation on images of a current frame and previous and next frames by using three frames of image information.
3. The unmanned aerial vehicle and artificial intelligence based violation building data identification method as recited in claim 2, wherein: the improved interframe difference method formula is as follows:
Figure 178572DEST_PATH_IMAGE001
；
Figure 73716DEST_PATH_IMAGE002
and
Figure 794809DEST_PATH_IMAGE003
the images are respectively the difference between the front frame and the back frame and the current frame,
Figure 244245DEST_PATH_IMAGE004
and
Figure 941943DEST_PATH_IMAGE005
、
Figure 393391DEST_PATH_IMAGE006
the gray values of the image at time t, time t-1 and time t +1 respectively,
Figure 49500DEST_PATH_IMAGE007
is a coefficient of gray scale to be used,Nis the total number of pixels of the region to be detected,
Figure 986232DEST_PATH_IMAGE008
an image to be detected is obtained;
and then, performing binarization processing on the frame difference image obtained by threshold T processing:
Figure 723506DEST_PATH_IMAGE009
；
wherein,
Figure 593242DEST_PATH_IMAGE010
or
Figure 623515DEST_PATH_IMAGE011
，
Figure 274639DEST_PATH_IMAGE012
Or both results from the binarization.
4. The unmanned aerial vehicle and artificial intelligence based violation building data identification method as recited in claim 1, wherein: in the S1.3, the preprocessing method comprises image gray processing, color inversion and a canny edge detection algorithm;
the picture gray level processing formula is as follows:
Figure 314140DEST_PATH_IMAGE013
；
wherein,
Figure 38382DEST_PATH_IMAGE015
、
Figure 6600DEST_PATH_IMAGE017
and
Figure 449083DEST_PATH_IMAGE019
respectively representing components of red, green and blue, and averaging the three components of the color image to obtain a gray value;
the color inversion formula is as follows:
Figure 761116DEST_PATH_IMAGE020
；
wherein,
Figure 307241DEST_PATH_IMAGE021
for the value of the current pixel, it is,
Figure 210475DEST_PATH_IMAGE022
is the pixel value after color inversion;
the canny edge detection algorithm is as follows:
Figure 609095DEST_PATH_IMAGE023
；
wherein,
Figure 226284DEST_PATH_IMAGE024
is the standard deviation of the Gaussian distribution,
Figure 128381DEST_PATH_IMAGE025
the pixel points are set;
multiplying each pixel point and the neighborhood thereof by a Gaussian matrix, and taking the weighted average value as the final gray value:
Figure 202516DEST_PATH_IMAGE026
；
Figure 327248DEST_PATH_IMAGE027
；
wherein,
Figure 246662DEST_PATH_IMAGE028
is the amplitude of the received signal and is,
Figure 268845DEST_PATH_IMAGE029
is the direction of the light beam emitted by the light source,
Figure 513881DEST_PATH_IMAGE030
and
Figure 122979DEST_PATH_IMAGE031
respectively the image at a pixel point
Figure 846085DEST_PATH_IMAGE032
Horizontal gradient magnitude and vertical gradient magnitude.
5. The unmanned aerial vehicle and artificial intelligence based violation building data identification method as recited in claim 4, wherein: after the canny edge detection algorithm is detected, calculating a gradient value and a gradient direction through a sobel operator, and filtering out a maximum value; setting eight gradient directions, including: 0 °,45 °,90 °,135 °,180 °,225 °,270 °,315 °;
and setting a threshold value:
Figure 988353DEST_PATH_IMAGE033
；
wherein,
Figure 902826DEST_PATH_IMAGE034
gray values of all points to be measured;
and filtering the noise according to the image definition evaluation value:
Figure 232176DEST_PATH_IMAGE035
；
wherein,
Figure 24552DEST_PATH_IMAGE036
is effective.
6. The violation building data identification method based on unmanned aerial vehicle and artificial intelligence as recited in claim 1, wherein: in S1.4, the artificial intelligence algorithm includes: image input, basic feature extraction, multi-layer complex feature extraction, feature learning and classification detection results; the method comprises the steps of carrying out convolution calculation on an image, outputting a classification result by a Softmax classifier through a pooling layer, an activation function and a full connection layer, and realizing the identification and judgment of the garbage.
7. The unmanned aerial vehicle and artificial intelligence based violation building data identification method as recited in claim 6, wherein: the convolution calculation is that a small matrix slides on an image or an input characteristic diagram, the output characteristic diagram is a result obtained by multiplication and addition of the matrix, and the calculation method is as follows:
Figure 991633DEST_PATH_IMAGE037
；
wherein,
Figure 578472DEST_PATH_IMAGE038
is the number of the convolution kernel channels, sum is the matrix addition operator,bis a characteristic parameter;
performing aggregation statistics through the features of different positions, and selecting a representative value to represent the original feature; by adopting the maxpololing method, the calculation formula is as follows:
Figure 395119DEST_PATH_IMAGE039
；
wherein,
Figure 241719DEST_PATH_IMAGE040
characteristic values of different positions;
introducing an activation function relu function into the neural network, wherein the calculation formula is as follows:
Figure 93000DEST_PATH_IMAGE041
；
wherein x is a characteristic value and is used for improving the characterization capability of the artificial intelligence algorithm.
8. The unmanned aerial vehicle and artificial intelligence based violation building data identification method of claim 6, wherein: the convolution layer, the pooling layer and the activation function structure map original data to a feature vector space, and the full connection layer is obtained by a calculation formula:
Figure 585162DEST_PATH_IMAGE042
；
wherein,
Figure 154683DEST_PATH_IMAGE043
and b, integrating and summarizing the learned distributed features for the mapped feature samples and the feature parameters, and then mapping the integrated distributed features to a sample mark space.
9. The unmanned aerial vehicle and artificial intelligence based violation building data identification method as recited in claim 6, wherein: softmax maps the feature vectors of the input neural network to (0, 1) space, the sum of these values is 1, and the maximum probability value is selected as the classification result:
Figure 259168DEST_PATH_IMAGE044
；
wherein,
Figure 230535DEST_PATH_IMAGE045
is a classification vector.
10. The unmanned aerial vehicle and artificial intelligence based violation building data identification method of claim 6, wherein: and after the artificial intelligence algorithm is processed, performing garbage detection, and summarizing detection results to terminal equipment for visual display and information storage.
CN202211219087.2A 2022-10-08 2022-10-08 Method for identifying and judging illegal garbage based on unmanned aerial vehicle and artificial intelligence Active CN115294486B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211219087.2A CN115294486B (en) 2022-10-08 2022-10-08 Method for identifying and judging illegal garbage based on unmanned aerial vehicle and artificial intelligence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211219087.2A CN115294486B (en) 2022-10-08 2022-10-08 Method for identifying and judging illegal garbage based on unmanned aerial vehicle and artificial intelligence

Publications (2)

Publication Number Publication Date
CN115294486A true CN115294486A (en) 2022-11-04
CN115294486B CN115294486B (en) 2023-01-13

Family

ID=83834617

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211219087.2A Active CN115294486B (en) 2022-10-08 2022-10-08 Method for identifying and judging illegal garbage based on unmanned aerial vehicle and artificial intelligence

Country Status (1)

Country Link
CN (1) CN115294486B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115588145A (en) * 2022-12-12 2023-01-10 深圳联和智慧科技有限公司 Unmanned aerial vehicle-based river channel garbage floating identification method and system

Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140118716A1 (en) * 2012-10-31 2014-05-01 Raytheon Company Video and lidar target detection and tracking system and method for segmenting moving targets
CN108230368A (en) * 2016-12-14 2018-06-29 贵港市瑞成科技有限公司 A kind of fast-moving target detection method
CN109460764A (en) * 2018-11-08 2019-03-12 中南大学 A kind of satellite video ship monitoring method of combination brightness and improvement frame differential method
CN109472200A (en) * 2018-09-29 2019-03-15 深圳市锦润防务科技有限公司 A kind of intelligent sea rubbish detection method, system and storage medium
CN110298323A (en) * 2019-07-02 2019-10-01 中国科学院自动化研究所 Detection method of fighting based on video analysis, system, device
CN110599523A (en) * 2019-09-10 2019-12-20 江南大学 ViBe ghost suppression method fused with interframe difference method
CN110910318A (en) * 2019-10-21 2020-03-24 中国科学院西安光学精密机械研究所 Weak contrast schlieren small ball center calculation method for comprehensive diagnosis light path quick automatic collimation system
CN111539296A (en) * 2020-04-17 2020-08-14 河海大学常州校区 Method and system for identifying illegal building based on remote sensing image change detection
CN112581492A (en) * 2019-09-27 2021-03-30 北京京东尚科信息技术有限公司 Moving target detection method and device
CN114120120A (en) * 2021-11-25 2022-03-01 广东电网有限责任公司 Method, device, equipment and medium for detecting illegal building based on remote sensing image
CN114241364A (en) * 2021-11-30 2022-03-25 南京理工大学 Method for quickly calibrating foreign object target of overhead transmission line

Patent Citations (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140118716A1 (en) * 2012-10-31 2014-05-01 Raytheon Company Video and lidar target detection and tracking system and method for segmenting moving targets
CN108230368A (en) * 2016-12-14 2018-06-29 贵港市瑞成科技有限公司 A kind of fast-moving target detection method
CN109472200A (en) * 2018-09-29 2019-03-15 深圳市锦润防务科技有限公司 A kind of intelligent sea rubbish detection method, system and storage medium
CN109460764A (en) * 2018-11-08 2019-03-12 中南大学 A kind of satellite video ship monitoring method of combination brightness and improvement frame differential method
CN110298323A (en) * 2019-07-02 2019-10-01 中国科学院自动化研究所 Detection method of fighting based on video analysis, system, device
CN110599523A (en) * 2019-09-10 2019-12-20 江南大学 ViBe ghost suppression method fused with interframe difference method
CN112581492A (en) * 2019-09-27 2021-03-30 北京京东尚科信息技术有限公司 Moving target detection method and device
CN110910318A (en) * 2019-10-21 2020-03-24 中国科学院西安光学精密机械研究所 Weak contrast schlieren small ball center calculation method for comprehensive diagnosis light path quick automatic collimation system
CN111539296A (en) * 2020-04-17 2020-08-14 河海大学常州校区 Method and system for identifying illegal building based on remote sensing image change detection
CN114120120A (en) * 2021-11-25 2022-03-01 广东电网有限责任公司 Method, device, equipment and medium for detecting illegal building based on remote sensing image
CN114241364A (en) * 2021-11-30 2022-03-25 南京理工大学 Method for quickly calibrating foreign object target of overhead transmission line

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
XIAOHE LUO等: "Improved Three-Frame-Difference Algorithm for Infrared Moving Target", 《IEEE》 *
赵柏山 等: "基于帧差与背景差分的改进目标识别算法", 《通信技术》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115588145A (en) * 2022-12-12 2023-01-10 深圳联和智慧科技有限公司 Unmanned aerial vehicle-based river channel garbage floating identification method and system

Also Published As

Publication number Publication date
CN115294486B (en) 2023-01-13

Similar Documents

Publication Publication Date Title
CN108615226B (en) Image defogging method based on generation type countermeasure network
Huang et al. An advanced single-image visibility restoration algorithm for real-world hazy scenes
CN103400150B (en) A kind of method and device that road edge identification is carried out based on mobile platform
Negru et al. Exponential contrast restoration in fog conditions for driving assistance
US9418402B2 (en) System for improving foggy luminance image using fog reduction estimation model
CN109918971B (en) Method and device for detecting number of people in monitoring video
CN103034983B (en) A kind of defogging method capable based on anisotropic filtering
CN112149543B (en) Building dust recognition system and method based on computer vision
CN103198459A (en) Haze image rapid haze removal method
CN106657948A (en) low illumination level Bayer image enhancing method and enhancing device
CN107636724A (en) The adaptive demister of wavelength and method of image
CN115294486B (en) Method for identifying and judging illegal garbage based on unmanned aerial vehicle and artificial intelligence
Sahu et al. Image dehazing based on luminance stretching
CN116052090A (en) Image quality evaluation method, model training method, device, equipment and medium
Khan et al. Recent advancement in haze removal approaches
Wang et al. Haze removal algorithm based on single-images with chromatic properties
Alami et al. Local fog detection based on saturation and RGB-correlation
CN104091306A (en) Image defogging method based on mathematic morphology and multi-resolution fusion
CN104299234B (en) The method and system that rain field removes in video data
Hu et al. A low illumination video enhancement algorithm based on the atmospheric physical model
Negru et al. Exponential image enhancement in daytime fog conditions
CN112598777B (en) Haze fusion method based on dark channel prior
Xiang et al. A new model for daytime visibility index estimation fused average sobel gradient and dark channel ratio
Roy et al. A ground truth annotated video dataset for moving object detection in degraded atmospheric outdoor scenes
Naseeba et al. KP Visibility Restoration of Single Hazy Images Captured in Real-World Weather Conditions

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
GR01 Patent grant
GR01 Patent grant