CN111539494A - Hydraulic protection damage detection method based on U-Net and SVM - Google Patents
Hydraulic protection damage detection method based on U-Net and SVM Download PDFInfo
- Publication number
- CN111539494A CN111539494A CN202010653611.1A CN202010653611A CN111539494A CN 111539494 A CN111539494 A CN 111539494A CN 202010653611 A CN202010653611 A CN 202010653611A CN 111539494 A CN111539494 A CN 111539494A
- Authority
- CN
- China
- Prior art keywords
- hydraulic protection
- net
- semantic segmentation
- svm
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Abstract
The invention belongs to the technical field of hydraulic protection facility inspection, and particularly relates to a hydraulic protection damage detection method based on U-Net and SVM, which comprises the following steps: s1, acquiring a field image containing a hydraulic protection facility in real time; s2, inputting the site image containing the hydraulic protection facility into the trained U-Net semantic segmentation network to extract the hydraulic protection facility image; and S3, inputting the image of the hydraulic protection facility into the trained SVM classifier to detect whether the hydraulic protection facility is damaged or not. The invention utilizes the U-Net semantic segmentation network to extract the hydraulic protection facilities from the site images which are collected in real time and contain the hydraulic protection facilities, thereby reducing the interference of background images on hydraulic protection damage detection; and then the data is input into an SVM classifier for recognition, whether the hydraulic protection facilities are damaged or not is detected, the data noise is low, and the detection precision is high.
Description
Technical Field
The invention belongs to the technical field of hydraulic protection facility inspection, and particularly relates to a hydraulic protection damage detection method based on U-Net and SVM.
Background
Hydraulic protection, which is used for preventing pipeline backfill soil loss and ambient environment damage caused by water flow scouring, so as to protect facilities such as a slope protection, a fort and the like for pipeline safety; mainly comprises retaining walls, revetments, drainage ditches, silt dams, water passing surfaces, piling of river-crossing pipelines, stone pressing cages, balancing weights, covering layers and the like.
The thickness of the soil covering of the pipeline and the water and soil protection state near the pipeline are always kept, the protection and the reinforcement are carried out, the water damage is timely remedied, the occurrence and the development of the water damage are prevented, and the safety of the pipeline can be effectively ensured. In particular to the construction of long-distance pipelines of petroleum and natural gas, because the passing regions and geological conditions of the pipelines are complex, such as unfavorable engineering geological sections of mountains, hills, deserts, swamps, rivers, mining areas and the like. The water damage problem threatens the pipeline seriously, and the hydraulic protection can reduce the influence of disasters such as water, landslide and the like generated by the water on the safety of the pipeline. Therefore, inspection of hydraulic protection facilities is particularly important.
At present, the inspection of hydraulic protection facilities still mainly depends on manual line inspection, the inspection cost is high, and the efficiency is low. In addition, the prior art also relates to intelligent detection technology of pipelines, such as: patent document No. CN110108783A discloses a method for detecting a pipeline defect based on a convolutional neural network, which includes: acquiring a defect sample of a test pipeline, and establishing an initial convolutional neural network model; training and learning the initial convolutional neural network model through a defect sample and obtaining a final convolutional neural network model for testing the size of the pipeline defect; and carrying out convolution neural processing on the circumferential magnetic flux leakage signal, the axial magnetic flux leakage signal and the radial magnetic flux leakage signal of the actual measurement pipeline based on the final convolution neural network model to obtain the defect size of the actual measurement pipeline. Namely, the defects are detected and identified by adopting a mode of combining feature extraction and a classifier.
However, the geographical environment of the hydraulic protection facility is complex, and the acquired background image in the image containing the hydraulic protection facility interferes with the damage detection of the hydraulic protection facility, so that the detection accuracy is not high.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a hydraulic protection damage detection method based on U-Net and SVM.
In order to achieve the purpose, the invention adopts the following technical scheme:
a hydraulic protection damage detection method based on U-Net and SVM comprises the following steps:
s1, acquiring a field image containing a hydraulic protection facility in real time;
s2, inputting the site image containing the hydraulic protection facility into the trained U-Net semantic segmentation network to extract the hydraulic protection facility image;
and S3, inputting the image of the hydraulic protection facility into the trained SVM classifier to detect whether the hydraulic protection facility is damaged or not.
As a preferred scheme, the training of the U-Net semantic segmentation network and the SVM classifier comprises the following steps:
SA, collecting normal samples and abnormal samples; the normal sample is an image containing normal hydraulic protection facilities, and the abnormal sample is an image containing damaged hydraulic protection facilities;
SB, carrying out image preprocessing on the normal sample and the abnormal sample, and respectively making a semantic segmentation data set and a hydraulic protection damage data set;
the SC utilizes the semantic segmentation data set to train the U-Net semantic segmentation network to obtain a trained U-Net semantic segmentation network; and training the SVM classifier by using the hydraulic protection damage data set to obtain the trained SVM classifier.
Preferably, in the step SB, the image preprocessing includes Z-score normalization and histogram equalization of the image.
Preferably, the step SB of creating the semantic segmentation data set includes:
labeling the normal sample and the abnormal sample after pretreatment by using LableMe, and dividing pixels in the image into a hydraulic protection facility and a non-hydraulic protection facility, wherein the hydraulic protection facility is labeled as 1, and the non-hydraulic protection facility is labeled as 0.
As a preferred scheme, in the step SC, training the U-Net semantic segmentation network by using a semantic segmentation dataset includes:
establishing a U-Net semantic segmentation network, wherein the U-Net semantic segmentation network comprises 4 times of pooling operation, 4 times of upsampling operation, 25 times of convolution operation and 4 times of cutting and copying operation;
and inputting the semantic segmentation data set into a U-Net semantic segmentation network for training.
Preferably, in the convolution operation, the convolution kernel size of 24 convolutional layers is 3 × 3, and the convolution kernel size of 1 convolutional layer is 1 × 1;
the pooling operation comprises 4 pooling layers with a pooling core size of 2 x 2, step size of 1, and padding of 0;
in the up-sampling operation, the sizes of deconvolution operation filters are all 3 × 3, the step length is 2, and the padding is 0.
As a preferred scheme, batch normalization processing is accessed after convolution operation with a convolution kernel size of 3 × 3.
Preferably, Dropout is added after the pooling layer of the encoding part, and Dropout processing is added in the middle of deconvolution and convolution of the de-encoding part;
and the hidden layer of the U-Net semantic segmentation network adopts a modified linear unit ReLU as an activation function, the output layer adopts a Sigmoid function as an activation function, and the output result is mapped into a [0, 1] interval.
Preferably, the training of the SVM classifier includes the following steps:
hydraulic protection damage feature extraction from hydraulic protection damage data set by utilizing SIFT algorithmx i ;
Constructing a training data setQ;
Wherein the content of the first and second substances,i=1, 2, …, n; n is the number of normal samples and abnormal samples,y i is a class label; when in usey i When the value is 1, representing that the hydraulic protection facility is damaged in the image; when in usey i When the value is-1, the representative image has no damage of hydraulic protection facilities;
the optimal hyperplane is as follows: (w-x+b)=0
In order to construct an optimal hyperplane, the maximum interval separation is converted into an optimization problem, and the optimization problem is converted into a Lagmage dual problem:
solving for b and w of the hyperplane:
Preferably, if the output result of the SVM classifier is that the hydraulic protection facility is damaged, an alarm is executed.
Compared with the prior art, the invention has the beneficial effects that:
the invention utilizes the U-Net semantic segmentation network to extract the hydraulic protection facilities from the site images which are collected in real time and contain the hydraulic protection facilities, thereby reducing the interference of background images on hydraulic protection damage detection; and then the data is input into an SVM classifier for recognition, whether the hydraulic protection facilities are damaged or not is detected, the data noise is low, and the detection precision is high.
Drawings
Fig. 1 is a flow chart of a hydraulic protection damage detection method based on U-Net and SVM in embodiment 1 of the present invention.
FIG. 2 is a training flow chart of the U-Net semantic segmentation network and the SVM classifier according to embodiment 1 of the present invention.
Detailed Description
In order to more clearly illustrate the embodiments of the present invention, the following description will explain the embodiments of the present invention with reference to the accompanying drawings. It is obvious that the drawings in the following description are only some examples of the invention, and that for a person skilled in the art, other drawings and embodiments can be derived from them without inventive effort.
Example 1:
as shown in fig. 1, the hydraulic protection damage detection method based on U-Net and SVM of the present embodiment includes the following steps:
s1, acquiring a field image containing a hydraulic protection facility in real time;
s2, inputting the site image containing the hydraulic protection facility into the trained U-Net semantic segmentation network to extract the hydraulic protection facility image;
and S3, inputting the image of the hydraulic protection facility into the trained SVM classifier to detect whether the hydraulic protection facility is damaged or not.
In step S1, the on-site image of the hydraulic protection facility is obtained by aerial photography by the unmanned aerial vehicle.
As shown in fig. 2, the training of the U-Net semantic segmentation network and the SVM classifier of the present embodiment includes the following steps:
SA, collecting normal samples and abnormal samples; the normal sample is an image containing normal hydraulic protection facilities, and the abnormal sample is an image containing damaged hydraulic protection facilities;
specifically, acquiring a site sample image of hydraulic protection by unmanned aerial vehicle aerial photography, and acquiring 5000 images around the hydraulic protection, wherein the size of the images is 1024 × 1024, 2500 images comprise normal hydraulic protection facilities, and 2500 photos comprise damage of the hydraulic protection facilities; the collected images need to contain images of various hydraulic protection facilities and images of various representative interferences, such as tree shadows, mountain shelters, building interferences and the like.
SB, carrying out image preprocessing on the normal sample and the abnormal sample, and respectively making a semantic segmentation data set and a hydraulic protection damage data set;
wherein the image pre-processing comprises Z-score normalization and histogram equalization of the image. Z-score standardization is carried out on the image, so that the processed image data are in accordance with standard normal distribution; since contrast information plays an important role in image segmentation, histogram equalization is employed to improve the contrast of grayscale images.
The semantic segmentation data set production of the embodiment includes:
and labeling the normal sample and the abnormal sample of the preprocessed on-site sample image aerial photographed by the unmanned aerial vehicle by using LableMe software, and dividing pixels in the image into a hydraulic protection facility and a non-hydraulic protection facility, wherein the hydraulic protection facility is labeled as 1, and the non-hydraulic protection facility is labeled as 0.
The manufacturing of the hydraulic protection damage data set of the embodiment includes:
and (4) segmenting an image with hydraulic protection facility damage and an image with normal hydraulic protection facility from the field sample image of the unmanned aerial vehicle aerial photograph after image preprocessing to manufacture a hydraulic protection damage data set.
The SC utilizes the semantic segmentation data set to train the U-Net semantic segmentation network to obtain a trained U-Net semantic segmentation network; and training the SVM classifier by using the hydraulic protection damage data set to obtain the trained SVM classifier.
In this embodiment, training a U-Net semantic segmentation network by using a semantic segmentation dataset includes:
firstly, establishing a U-Net semantic segmentation network, wherein the U-Net semantic segmentation network comprises 4 times of pooling operation, 4 times of upsampling operation, 25 times of convolution operation, and 4 times of cutting and copying operation; in the convolution operation, the convolution kernel size of 24 convolution layers is 3 × 3, and the convolution kernel size of 1 convolution layer is 1 × 1; the pooling operation comprised 4 pooling layers with a pooling core size of 2 × 2, step size of 1, and padding of 0; in the upsampling operation, the sizes of the deconvolution operation filters are all 3 × 3, the step size is 2, and the padding is 0.
In addition, batch normalization processing is accessed after convolution operation with the convolution kernel size of 3 x 3 so as to adapt to data with different distributions and achieve the effect of improving the model training speed and the generalization capability. Wherein, the batch normalization processing steps are as follows:
wherein the content of the first and second substances,Mfor all the pixel numbers of the feature map after the convolution operation,for each of the sample points, the number of the sample points,is the average value of all the pixel points,as a function of the variance of the batch of data,is a very small positive number, to ensure that the denominator is greater than 0,are sample points after batch normalization.
In order to eliminate the over-fitting problem of model training, Dropout is added after the pooling layer of the encoding part, and Dropout processing is added in the middle of deconvolution and convolution of the decoding part.
The hidden layer of the built U-Net semantic segmentation network adopts a modified linear unit ReLU as an activation function, and the function of the hidden layer enables a model to approach any function in a nonlinear mode, so that the expression capability of a deep neural network is improved.
And the output layer of the established U-Net semantic segmentation network adopts a Sigmoid function as an activation function, and an output result is mapped into a [0, 1] interval for secondary classification.
The loss function of the built U-Net semantic segmentation network adopts binary cross entropy as follows:
wherein the content of the first and second substances,in order to be the real output,is the actual output of the neuron.
The built U-Net semantic segmentation network adopts a Momentum optimization method to improve the learning speed and the convergence speed of the model.
The embodiment trains the SVM classifier by utilizing the hydraulic protection damage data set, and comprises the following steps:
hydraulic protection damage feature extraction from hydraulic protection damage data set by utilizing SIFT algorithmx i ;
Constructing a training data setQ;
Wherein the content of the first and second substances,i=1, 2, …, n; n is the number of normal samples and abnormal samples,y i is a class label; when in usey i When the value is 1, representing that the hydraulic protection facility is damaged in the image; when in usey i When the value is-1, the representative image has no damage of hydraulic protection facilities;
the optimal hyperplane is as follows: (w-x+b)=0
In order to construct an optimal hyperplane, the maximum interval separation is converted into an optimization problem, and the optimization problem is converted into a Lagmage dual problem:
solving for b and w of the hyperplane:
And detecting the hydraulic protection facilities after semantic segmentation by using the trained SVM classifier, and sending an alarm signal to inform line patrol personnel to arrive at a hydraulic protection damage site for protection if the hydraulic protection facilities are detected to be damaged.
Example 2:
the hydraulic protection damage detection method based on U-Net and SVM of the embodiment is different from the embodiment 1 in that:
in step S1, after the live images containing the hydraulic protection facilities are collected in real time, before the trained U-Net semantic segmentation network is input, histogram equalization is performed on the live images containing the hydraulic protection facilities collected in real time, so as to improve the contrast of the grayscale images and improve the detection effect and accuracy of the hydraulic protection facilities.
Other steps can be referred to example 1.
Example 3:
the hydraulic protection damage detection method based on U-Net and SVM of the embodiment is different from the embodiment 1 in that:
the image acquisition of the hydraulic protection facility is not limited to the unmanned aerial vehicle acquisition of embodiment 1, and may be performed by image acquisition equipment such as a camera along the hydraulic protection facility.
Other steps can be referred to example 1.
The foregoing has outlined rather broadly the preferred embodiments and principles of the present invention and it will be appreciated that those skilled in the art may devise variations of the present invention that are within the spirit and scope of the appended claims.
Claims (10)
1. The hydraulic protection damage detection method based on the U-Net and the SVM is characterized by comprising the following steps of:
s1, acquiring a field image containing a hydraulic protection facility in real time;
s2, inputting the site image containing the hydraulic protection facility into the trained U-Net semantic segmentation network to extract the hydraulic protection facility image;
and S3, inputting the image of the hydraulic protection facility into the trained SVM classifier to detect whether the hydraulic protection facility is damaged or not.
2. The hydraulic protection damage detection method based on U-Net and SVM of claim 1, wherein the training of the U-Net semantic segmentation network and SVM classifier comprises the following steps:
SA, collecting normal samples and abnormal samples; the normal sample is an image containing normal hydraulic protection facilities, and the abnormal sample is an image containing damaged hydraulic protection facilities;
SB, carrying out image preprocessing on the normal sample and the abnormal sample, and respectively making a semantic segmentation data set and a hydraulic protection damage data set;
the SC utilizes the semantic segmentation data set to train the U-Net semantic segmentation network to obtain a trained U-Net semantic segmentation network; and training the SVM classifier by using the hydraulic protection damage data set to obtain the trained SVM classifier.
3. The hydraulic protection damage detection method based on U-Net and SVM of claim 2, wherein in the step SB, the image preprocessing comprises Z-score normalization and histogram equalization of the image.
4. The hydraulic protection damage detection method based on U-Net and SVM of claim 2, wherein in the step SB, the making of the semantic segmentation data set comprises:
labeling the normal sample and the abnormal sample after pretreatment by using LableMe, and dividing pixels in the image into a hydraulic protection facility and a non-hydraulic protection facility, wherein the hydraulic protection facility is labeled as 1, and the non-hydraulic protection facility is labeled as 0.
5. The hydraulic protection damage detection method based on U-Net and SVM of claim 4, wherein in the step SC, training the U-Net semantic segmentation network by using the semantic segmentation data set comprises:
establishing a U-Net semantic segmentation network, wherein the U-Net semantic segmentation network comprises 4 times of pooling operation, 4 times of upsampling operation, 25 times of convolution operation and 4 times of cutting and copying operation;
and inputting the semantic segmentation data set into a U-Net semantic segmentation network for training.
6. The hydraulic protection damage detection method based on U-Net and SVM of claim 5, wherein the convolution operation has convolution kernel size of 3 x 3 for 24 convolution layers and convolution kernel size of 1 x 1 for 1 convolution layer;
the pooling operation comprises 4 pooling layers with a pooling core size of 2 x 2, step size of 1, and padding of 0;
in the up-sampling operation, the sizes of deconvolution operation filters are all 3 × 3, the step length is 2, and the padding is 0.
7. The hydraulic protection damage detection method based on U-Net and SVM of claim 6, wherein the convolution operation with convolution kernel size of 3 x 3 is followed by batch normalization processing.
8. The hydraulic protection damage detection method based on U-Net and SVM of claim 6, wherein Dropout is added after the pooling layer of the encoding part, and Dropout processing is added in the middle of deconvolution and convolution of the de-encoding part;
and the hidden layer of the U-Net semantic segmentation network adopts a modified linear unit ReLU as an activation function, the output layer adopts a Sigmoid function as an activation function, and the output result is mapped into a [0, 1] interval.
9. The hydraulic protection damage detection method based on U-Net and SVM of claim 2, wherein the training of the SVM classifier comprises the following steps:
hydraulic protection damage feature extraction from hydraulic protection damage data set by utilizing SIFT algorithmx i ;
Constructing a training data setQ;
Wherein the content of the first and second substances,i=1, 2, …, n; n is normalThe number of samples and the number of abnormal samples,y i is a class label; when in usey i When the value is 1, representing that the hydraulic protection facility is damaged in the image; when in usey i When the value is-1, the representative image has no damage of hydraulic protection facilities;
the optimal hyperplane is as follows: (w-x+b)=0;
In order to construct an optimal hyperplane, the maximum interval separation is converted into an optimization problem, and the optimization problem is converted into a Lagmage dual problem:
solving for b and w of the hyperplane:
10. The hydraulic protection damage detection method based on U-Net and SVM of any one of claims 1-9, wherein if the output result of the SVM classifier is that there is damage to the hydraulic protection facility, an alarm is executed.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010653611.1A CN111539494A (en) | 2020-07-08 | 2020-07-08 | Hydraulic protection damage detection method based on U-Net and SVM |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010653611.1A CN111539494A (en) | 2020-07-08 | 2020-07-08 | Hydraulic protection damage detection method based on U-Net and SVM |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111539494A true CN111539494A (en) | 2020-08-14 |
Family
ID=71979837
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010653611.1A Pending CN111539494A (en) | 2020-07-08 | 2020-07-08 | Hydraulic protection damage detection method based on U-Net and SVM |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111539494A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113298846A (en) * | 2020-11-18 | 2021-08-24 | 西北工业大学 | Intelligent interference detection method based on time-frequency semantic perception |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110245673A (en) * | 2018-03-09 | 2019-09-17 | 浙江宇视科技有限公司 | Method for detecting parking stalls and device |
CN110321897A (en) * | 2019-07-08 | 2019-10-11 | 四川九洲视讯科技有限责任公司 | Divide the method for identification non-motor vehicle abnormal behaviour based on image, semantic |
CN110717886A (en) * | 2019-09-03 | 2020-01-21 | 南京理工大学 | Pavement pool detection method based on machine vision in complex environment |
CN111242955A (en) * | 2020-01-21 | 2020-06-05 | 河海大学 | Road surface crack image segmentation method based on full convolution neural network |
CN111339823A (en) * | 2019-12-30 | 2020-06-26 | 南京艾特斯科技有限公司 | Threshing and sunning ground detection method based on machine vision and back projection algorithm |
-
2020
- 2020-07-08 CN CN202010653611.1A patent/CN111539494A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110245673A (en) * | 2018-03-09 | 2019-09-17 | 浙江宇视科技有限公司 | Method for detecting parking stalls and device |
CN110321897A (en) * | 2019-07-08 | 2019-10-11 | 四川九洲视讯科技有限责任公司 | Divide the method for identification non-motor vehicle abnormal behaviour based on image, semantic |
CN110717886A (en) * | 2019-09-03 | 2020-01-21 | 南京理工大学 | Pavement pool detection method based on machine vision in complex environment |
CN111339823A (en) * | 2019-12-30 | 2020-06-26 | 南京艾特斯科技有限公司 | Threshing and sunning ground detection method based on machine vision and back projection algorithm |
CN111242955A (en) * | 2020-01-21 | 2020-06-05 | 河海大学 | Road surface crack image segmentation method based on full convolution neural network |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113298846A (en) * | 2020-11-18 | 2021-08-24 | 西北工业大学 | Intelligent interference detection method based on time-frequency semantic perception |
CN113298846B (en) * | 2020-11-18 | 2024-02-09 | 西北工业大学 | Interference intelligent detection method based on time-frequency semantic perception |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109934200B (en) | RGB color remote sensing image cloud detection method and system based on improved M-Net | |
Kang et al. | Robust median filtering forensics using an autoregressive model | |
CN111274918B (en) | River dryout and cutoff monitoring method and device based on multi-source remote sensing image | |
Dotel et al. | Disaster assessment from satellite imagery by analysing topographical features using deep learning | |
Noh et al. | Unsupervised change detection based on image reconstruction loss | |
Kaur et al. | A review on natural disaster detection in social media and satellite imagery using machine learning and deep learning | |
CN111539494A (en) | Hydraulic protection damage detection method based on U-Net and SVM | |
CN113469097B (en) | Multi-camera real-time detection method for water surface floaters based on SSD network | |
CN117078943B (en) | Remote sensing image road segmentation method integrating multi-scale features and double-attention mechanism | |
CN114241310A (en) | Improved YOLO model-based intelligent identification method for piping dangerous case of dike | |
CN113887472A (en) | Remote sensing image cloud detection method based on cascade color and texture feature attention | |
CN113158770A (en) | Improved mining area change detection method of full convolution twin neural network | |
Meng et al. | A modified fully convolutional network for crack damage identification compared with conventional methods | |
CN112800851A (en) | Water body contour automatic extraction method and system based on full convolution neuron network | |
CN116843946A (en) | Tunnel rock mass main structural surface identification method and device based on image identification | |
CN115456957B (en) | Method for detecting change of remote sensing image by full-scale feature aggregation | |
Khan et al. | Analyzing the Land Cover Change and Degradation in Sundarbans Mangrove Forest Using Machine Learning and Remote Sensing Technique | |
CN114724023A (en) | Twin network-based water body change detection method | |
CN113963271A (en) | Model for identifying impervious surface from remote sensing image and method for training model | |
Yang et al. | Cadnet: top-down contextual saliency detection network for high spatial resolution remote sensing image shadow detection | |
Kang et al. | Stochastic properties of coastal flooding events–Part 1: convolutional-neural-network-based semantic segmentation for water detection | |
Miao et al. | Deep learning landslide extraction based on multi-temporal image difference local spatial information | |
Zhao et al. | SqUNet: An High-performance Network for Crater Detection with DEM data | |
Carani et al. | Detection of Tornado damage in forested regions via convolutional neural networks and uncrewed aerial system photogrammetry | |
CN116091907B (en) | Image tampering positioning model and method based on non-mutually exclusive ternary comparison learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200814 |
|
RJ01 | Rejection of invention patent application after publication |