CN112906459A - Road network checking technology based on high-resolution remote sensing image and deep learning method - Google Patents
Road network checking technology based on high-resolution remote sensing image and deep learning method Download PDFInfo
- Publication number
- CN112906459A CN112906459A CN202110031846.1A CN202110031846A CN112906459A CN 112906459 A CN112906459 A CN 112906459A CN 202110031846 A CN202110031846 A CN 202110031846A CN 112906459 A CN112906459 A CN 112906459A
- Authority
- CN
- China
- Prior art keywords
- road network
- image
- objective function
- deep learning
- road
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Software Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Probability & Statistics with Applications (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Medical Informatics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a road network checking technology based on a high-resolution remote sensing image and a deep learning method, which belongs to the technical field of road network checking and comprises four steps of sample set manufacturing, model training, road network extraction and road network checking; the data set used for preparing the sample set is derived from a high-resolution second satellite image; the model training comprises setting an objective function, optimizing the objective function and performing iterative training; the road network extracts a road network grid binary image, vectorizes the road network grid binary image and outputs a road network vector result; and the road network check judges the spatial matching relationship of the two by using a buffer area analysis method. The method greatly reduces the cost of manpower and material resources, has more advantages in extraction efficiency, extraction precision and universality, and is suitable for various landform environments.
Description
Technical Field
The invention belongs to the technical field of road network checking, and particularly relates to a road network checking technology based on a high-resolution remote sensing image and a deep learning method.
Background
The large-scale construction and upgrading transformation of the road network greatly promote the development of the traffic industry and the economy of China, but also put forward higher requirements on the comprehensive supervision of the road network. Especially for road network inspection, the traditional manual field inspection is time-consuming and labor-consuming, the data acquisition period is too long, and the current actual work needs cannot be met, so that a more intelligent and automatic technical means is urgently needed.
The basis of road network verification is road network extraction. High-resolution (high-resolution) remote sensing is used as an advanced earth observation means, can comprehensively, quickly, accurately and objectively acquire images of large-range ground targets, and is very suitable for road network extraction work. Scholars at home and abroad propose various road network automatic extraction algorithms based on high-resolution remote sensing, so that manpower and material resources are greatly saved, but the problems of low accuracy, low extraction efficiency, weak generalization capability and the like still exist. In recent years, the rapid development of artificial intelligence technology enables the deep learning method to obtain remarkable results in the fields of image recognition, detection, tracking and the like, and provides new ideas and inspiration for road network extraction.
There are various road network extraction methods based on high-resolution remote sensing. Traditionally, a manual interpretation method is applied, namely, a road contour is sketched on a high-resolution remote sensing image through a manual target. Although the manual interpretation has high extraction precision and small technical difficulty, the manual interpretation has the disadvantages of high cost and low efficiency when facing a large-scale road network extraction task. With the performance improvement of computing equipment and the deep research of road network extraction algorithms, domestic and foreign scholars put forward various road network automatic extraction algorithms based on high-resolution remote sensing. Most of these methods are based on classification models, so they can be broadly classified into supervised classification methods and unsupervised classification methods. The supervised classification methods are currently mainstream methods, and include an artificial neural network method of Mnih and Hinton (2010), an object-oriented method of Huang et al (2009) based on a support vector machine, and a markov random field method of Zhu et al (2011). Unsupervised classification methods include the k-means clustering method of Maurya et al (2011), the mean shift clustering method of Miao et al (2014), and the probabilistic and graph-theoretic methods of Unsalan and sirnacek (2012). Other classical methods also include the snake method of Laptev et al (2000) and the higher order CRF method of Wegner et al (2015). Compared with a manual interpretation method, the methods greatly save manpower and material resources, and simultaneously keep higher accuracy, so that automation of road network extraction becomes possible.
The current mainstream road network extraction method based on high-resolution remote sensing images comprises a manual interpretation and supervision classification method. The accuracy of manual interpretation is high, but the extraction efficiency is low. The traditional supervision classification method can preliminarily realize automatic extraction. However, the extraction accuracy is low, a large amount of feature engineering work is needed in the early stage of extraction, the extraction accuracy is low, manual post-processing work is more, and the overall extraction process speed is low. In addition, the migration effect in areas with large range and complex landform conditions is poor.
Although the existing supervised classification road network extraction method effectively reduces the cost of manual extraction and keeps higher extraction accuracy, the following defects still exist:
1) under the influence of road surface shelters such as trees, shadow, the road network extraction accuracy still can not reach the requirement of practical application.
2) A large amount of feature engineering work is needed in the early stage of road network extraction work, and a large amount of manual post-processing work is needed due to the fact that the extraction accuracy rate is low, so that the speed of the whole extraction process is low, and the automation degree is low.
3) The existing method has a good extraction effect in a small range and a specific area, but has a poor migration effect in a large range and an area with complex topographic and geomorphic conditions.
SUMMERY OF THE UTILITY MODEL
The invention aims to provide a road network checking technology based on a high-resolution remote sensing image and a deep learning method so as to solve the problems in the background technology.
In order to achieve the purpose, the invention provides the following technical scheme:
the road network checking technology based on the high-resolution remote sensing image and the deep learning method comprises four steps of sample set manufacturing, model training, road network extraction and road network checking, and comprises the following specific operation steps:
the method comprises the following steps: preparing a sample set: the data set used for preparing the sample set is derived from a high-resolution second satellite image; the cloud coverage of all images is less than 5%; and preprocessing the acquired high-resolution second image, wherein the preprocessing comprises radiation correction, geometric correction and multiband fusion, and the fused high-resolution second image has four wavebands.
Step two: model training: the used deep learning model is Wasserstein GAN, a generator G in the WGAN model is composed of a general semantic segmentation network, a discriminator D is composed of a simple residual error network, and after judging whether data come from a generated sample G (x, z) or a real sample y, the probability p that the data are real samples is output; the method comprises the following specific steps:
s1, setting an objective function;
s2, optimizing an objective function;
s3, iterative training;
step three: road network extraction: and cutting the high-resolution second image for testing into 512 x 512 image slices, inputting the image slices into the generator G obtained in the step two, extracting a road network grid binary image, carrying out vectorization on the road network grid binary image, and finally outputting a road network vector result.
Step four: road network checking: and (4) converting the road network result extracted in the third step and the existing road network data into a unified coordinate system, and judging the spatial matching relationship of the road network result and the existing road network data by using a buffer area analysis method, so that preliminary road network inspection can be realized.
As a further scheme of the invention: in the first step, the road value and the non-road value in the binary raster image generated by the sample set preparation are 1 and 0 respectively.
As a further scheme of the invention: and (5) if the probability output value of the real sample in the model training of the step two is p >0.5, outputting 1, and otherwise, outputting 0.
As a further scheme of the invention: the step two model training is in the traditional LL1(G) A spatial penalty term ω is added, as shown in formula (3):
if pixel j is a way, ω is set to 1; if pixel j is not a way, ω is set to α, α > 1.
As a further scheme of the invention: the three types of road network check are check of the existing road network line type error and deviation, check of the existing road network line type change and check of the newly added road network line type.
Compared with the prior art, the method has the advantages that the cost of manpower and material resources is greatly reduced, the extraction efficiency, the extraction precision and the universality are superior, and the method is suitable for various landform environments and various landform environments; by judging the spatial matching relationship between the road network extraction result and the existing road network data, objective and accurate road network checking data can be obtained, and data phenomena such as false reporting, hidden reporting, missed reporting and the like are reduced; the deep learning method is introduced into road network extraction, high-resolution remote sensing is combined with a WGAN deep learning model, the accuracy, the speed and the migration capability of road network extraction are further improved, and technical support is provided for road network inspection; by judging the spatial matching relationship between the road network extraction result and the existing road network data, the verification of the existing road network line type errors, the deviation, the lane change and the newly added road network line type can be preliminarily realized.
Drawings
Fig. 1 is a distribution diagram of a road network inspection technology sample data set based on a high-resolution remote sensing image and a deep learning method.
FIG. 2 is a labeled sample set diagram of road network checking technology binaryzation based on a high-resolution remote sensing image and a deep learning method.
Fig. 3 is a model structure diagram of WGAN in the road network verification technology based on a high-resolution remote sensing image and a deep learning method.
FIG. 4 is a road network checking technology road network extraction result overall effect graph based on a high-resolution remote sensing image and a deep learning method.
FIG. 5 is a road network linear error checking graph of the prior art based on the road network checking technology of the high-resolution remote sensing image and the deep learning method.
FIG. 6 is a road network linear missing checking graph in the prior art of road network checking based on high-resolution remote sensing images and deep learning method.
Detailed Description
The technical solution of the present patent will be described in further detail with reference to the following embodiments.
Referring to fig. 1-6, the road network verification technique based on the high-resolution remote sensing image and the deep learning method includes four steps of sample set preparation, model training, road network extraction and road network verification, and the specific operation steps are as follows:
the method comprises the following steps: preparing a sample set: the data set used for manufacturing the sample set is derived from a high-resolution second satellite image, the high-resolution second satellite is a first civil optical remote sensing satellite independently developed by China, the spatial resolution of the first civil optical remote sensing satellite is superior to 1 m, a 1 m-resolution panchromatic camera and a 4 m-resolution multispectral camera are carried, and the high-resolution second satellite has the characteristics of high precision, long service life, multiple angles and the like; the cloud coverage of all images is less than 5%, and the images are clear, so that the complete coverage area and reliable quality of the subsequent road network extraction result are ensured; in addition, the acquired high-resolution second-number image is preprocessed, wherein the preprocessing comprises radiation correction, geometric correction and multiband fusion, the fused high-resolution second-number image has four wavebands (blue, green, red and near infrared wavebands), and the space resolution of the sub-satellite point can reach 0.8 meter.
As shown in fig. 1 and 2, before the model training is performed, the data set needs to be cut and labeled manually, and the specific steps are as follows:
s1, cutting the high-resolution second image into 512 x 512 image slices, and selecting 2 ten thousand slices which cover various landforms and contain road elements as a training sample set x;
s2, for each slice in the training sample set x, drawing a road line vector in ArcGIS software;
s3, a binarized raster image y of the corresponding slice is generated from the line vector, in which the road (positive sample) value is 1 and the non-road (negative sample) value is 0.
Step two: model training: the GAN model was proposed by Goodfellow in 2014, which requires training a Generator (Generator, abbreviated as G) that generates a generated sample approximating a real sample from random noise, and a Discriminator (Discriminator, abbreviated as D) that judges whether input data is from the real sample or the generated sample; the final training goal is to have G the ability to falsely and falsely, and D cannot distinguish whether the input data is a real sample or a generated sample; the deep learning model used by the invention is Wasserstein GAN (WGAN), which can effectively solve the defects of gradient disappearance, mode collapse and the like in the training process of a general GAN generator, and the model structure is shown in FIG. 3.
The generator G in the WGAN model is composed of a universal semantic segmentation network (such as U-Net), a high-resolution image slice sample set x with the size of 512 multiplied by 512 and random noise z are input into the generator, a road element generation sample G (x, z) generated according to original image characteristics is obtained, the discriminator D is composed of a simple residual error network, after judging whether the data is from the generation sample G (x, z) or a real sample y, the probability p that the output data is a real sample is output, if the probability p is greater than 0.5, 1 is output, and if the probability p is greater than 0.5, 0 is output; the method comprises the following specific steps:
s1, setting an objective function: overall objective function Ltotal(G, D) the objective function used with reference to the pix2pix model, as shown in equation (1):
LcGAN(G,D)+βLL1(G) (1)
wherein L iscGAN(G, D) is the objective function of cGAN, LL1(G) Is an objective function of the pixel pair, beta is a weight coefficient, LL1(G) As shown in formula (2):
Ex,z,y[ω||G(x,z)-y||1] (2)
the problem of imbalance of positive and negative samples caused by sparse distribution of a road network in a high-resolution image is solved in the conventional L methodL1(G) A spatial penalty term ω is added, as shown in formula (3):
if pixel j is a way, ω is set to 1; if pixel j is not a way, ω is set to α (α > 1). L iscGAN(G, D) is represented by the formula (4):
Ex,y[log D(x,y)]Ex,z[log(1(D(x,G(x,z)))] (4)
conventional LcGAN(G, D) evaluating the gap between the real and generated samples using JS divergence, which may lead to problems of "gradient vanishing", "mode collapse", etc., therefore the modified L is substituted for the traditional JS divergence by the Wasserstein distancecGAN(G, D) is represented by the formula (5):
whereinRepresentsThe sampling is from the connected space formed by the real sample distribution y and the generated sample distribution G (x, z). Lambda is a penalty factor which is the sum of,representsAboutOf the gradient of (c).
S2, optimizing an objective function: inputting the training sample set into a generator and a discriminator of the WGAN network, and calculating an objective function L according to an equation (1)total(G, D), as shown in formula (6), optimizing the value of the objective function, and updating the network parameters of the generator and the discriminator respectively, that is, fixing G first, maximizing D, updating the parameters of D, fixing D again, minimizing G, updating the parameters of G:
s3, iterative training: and repeating the training process until the optimization result is close to Nash equilibrium or the gradient of the loss function of the generator and the discriminator is almost not changed, and obtaining the optimal generator G.
Step three: road network extraction: as shown in fig. 4, the top-ranked second image for testing is cut into 512 × 512 image slices, and the image slices are input into the generator G obtained in step two, and a road network grid binary image is extracted and vectorized, and finally a road network vector result is output.
Step four: road network checking: as shown in fig. 5 and 6, the road network verification converts the road network result extracted in the third step and the existing road network data into a unified coordinate system, and determines the spatial matching relationship between the road network result and the existing road network data by using a buffer analysis method, so as to realize preliminary road network verification; currently, three types of verification are mainly implemented: the method comprises the steps of checking the existing road network line type errors and offsets, checking the existing road network line type changes (road reconstruction and extension), and checking the newly added road network line type (newly repaired roads or existing roads are missing).
Although the preferred embodiments of the present patent have been described in detail, the present patent is not limited to the above embodiments, and various changes can be made without departing from the spirit of the present patent within the knowledge of those skilled in the art.
Claims (5)
1. The road network checking technology based on the high-resolution remote sensing image and the deep learning method is characterized by comprising the following specific operation steps of:
the method comprises the following steps: preparing a sample set: and collecting satellite images, and preprocessing the acquired images, wherein the preprocessing comprises radiation correction, geometric correction and multiband fusion, and the fused images have four wavebands.
Before the model training is carried out, the data set needs to be cut and marked manually, and the method comprises the following specific steps:
s1, cutting the high-resolution second image into 512 x 512 image slices, and selecting 2 ten thousand slices which cover various landforms and contain road elements as a training sample set x;
s2, for each slice in the training sample set x, drawing a road line vector in ArcGIS software;
s3, generating a binary grid image y of the corresponding slice according to the line vector;
step two: model training: the used deep learning model is WasserteinGAN, namely WGAN, a generator G in the WGAN model is composed of a universal semantic segmentation network, a discriminator D is composed of a simple residual error network, and after judging whether data come from a generated sample G (x, z) or a real sample y, the probability p that the data are real samples is output; the method comprises the following specific steps:
s1, setting an objective function: overall objective function Ltotal(G, D) the objective function used with reference to the pix2pix model, as shown in equation (1):
LcGAN(G,D)+βLL1(G) (1)
wherein L iscGAN(G, D) is the objective function of cGAN, LL1(G) Is an objective function of the pixel pair, beta is a weight coefficient, LL1(G) As shown in formula (2):
Ex,z,y[ω||G(x,z)-y||1] (2)
LcGAN(G, D) is represented by the formula (4):
Ex,y[logD(x,y)]+Ex,z[log(1-(D(x,G(x,z)))] (4)
modified L Using Wasserstein distance instead of traditional JS divergence, modified LcGAN(G, D) is represented by the formula (5):
whereinRepresentsIn the connected space formed by the real sample distribution y and the generated sample distribution G (x, z), lambda is the penalty coefficient,representsAboutA gradient of (a);
s2, optimizing an objective function: inputting the training sample set into a generator and a discriminator of the WGAN network, and calculating an objective function L according to an equation (1)total(G, D), as shown in formula (6), optimizing the value of the objective function, and updating the network parameters of the generator and the discriminator respectively, that is, fixing G first, maximizing D, updating the parameters of D, fixing D again, minimizing G, updating the parameters of G:
s3, iterative training: repeating the training process until the optimization result is close to Nash equilibrium or the loss function gradient of the generator and the discriminator is almost not changed any more, and obtaining an optimal generator G;
step three: road network extraction: and cutting the high-resolution second image for testing into 512 x 512 image slices, inputting the image slices into the generator G obtained in the step two, extracting a road network grid binary image, carrying out vectorization on the road network grid binary image, and finally outputting a road network vector result.
Step four: road network checking: and (4) converting the road network result extracted in the third step and the existing road network data into a unified coordinate system, and judging the spatial matching relationship of the road network result and the existing road network data by using a buffer area analysis method, so that preliminary road network inspection can be realized.
2. The road network inspection technology based on the high-resolution remote sensing image and the deep learning method as claimed in claim 1, wherein the road value and the non-road value in the binarized raster image generated by the sample set of the first step are 1 and 0, respectively.
3. The road network verification technology based on the high-resolution remote sensing image and the deep learning method as claimed in claim 1, wherein the probability output value of the real sample in the model training of step two is 1 if p >0.5, otherwise 0 is output.
4. The road network verification technology based on the high-resolution remote sensing image and the deep learning method as claimed in claim 1, wherein the step two model training is performed in a conventional L modeL1(G) A spatial penalty term ω is added, as shown in formula (3):
if pixel j is a way, ω is set to 1; if pixel j is not a way, ω is set to α, α > 1.
5. The road network verification technology based on the high-resolution remote sensing image and the deep learning method as claimed in claim 1, wherein the three types of the road network verification are verification of an existing road network line type error and deviation, verification of an existing road network line type change and verification of a newly added road network line type.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110031846.1A CN112906459A (en) | 2021-01-11 | 2021-01-11 | Road network checking technology based on high-resolution remote sensing image and deep learning method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110031846.1A CN112906459A (en) | 2021-01-11 | 2021-01-11 | Road network checking technology based on high-resolution remote sensing image and deep learning method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN112906459A true CN112906459A (en) | 2021-06-04 |
Family
ID=76112338
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110031846.1A Pending CN112906459A (en) | 2021-01-11 | 2021-01-11 | Road network checking technology based on high-resolution remote sensing image and deep learning method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112906459A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118052869A (en) * | 2024-04-15 | 2024-05-17 | 深圳市峰和数智科技有限公司 | Unmanned aerial vehicle pose parameter optimization method and device, storage medium and computer equipment |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108256424A (en) * | 2017-12-11 | 2018-07-06 | 中交信息技术国家工程实验室有限公司 | A kind of high-resolution remote sensing image method for extracting roads based on deep learning |
CN109635748A (en) * | 2018-12-14 | 2019-04-16 | 中国公路工程咨询集团有限公司 | The extracting method of roadway characteristic in high resolution image |
CN109977922A (en) * | 2019-04-11 | 2019-07-05 | 电子科技大学 | A kind of pedestrian's mask generation method based on generation confrontation network |
CN110223259A (en) * | 2019-06-14 | 2019-09-10 | 华北电力大学(保定) | A kind of road traffic fuzzy image enhancement method based on production confrontation network |
CN110289927A (en) * | 2019-07-01 | 2019-09-27 | 上海大学 | The channel simulation implementation method of confrontation network is generated based on condition |
CN110443867A (en) * | 2019-08-01 | 2019-11-12 | 太原科技大学 | Based on the CT image super-resolution reconstructing method for generating confrontation network |
CN110569796A (en) * | 2019-09-09 | 2019-12-13 | 南京东控智能交通研究院有限公司 | Method for dynamically detecting lane line and fitting lane boundary |
CN111899172A (en) * | 2020-07-16 | 2020-11-06 | 武汉大学 | Vehicle target detection method oriented to remote sensing application scene |
-
2021
- 2021-01-11 CN CN202110031846.1A patent/CN112906459A/en active Pending
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108256424A (en) * | 2017-12-11 | 2018-07-06 | 中交信息技术国家工程实验室有限公司 | A kind of high-resolution remote sensing image method for extracting roads based on deep learning |
CN109635748A (en) * | 2018-12-14 | 2019-04-16 | 中国公路工程咨询集团有限公司 | The extracting method of roadway characteristic in high resolution image |
CN109977922A (en) * | 2019-04-11 | 2019-07-05 | 电子科技大学 | A kind of pedestrian's mask generation method based on generation confrontation network |
CN110223259A (en) * | 2019-06-14 | 2019-09-10 | 华北电力大学(保定) | A kind of road traffic fuzzy image enhancement method based on production confrontation network |
CN110289927A (en) * | 2019-07-01 | 2019-09-27 | 上海大学 | The channel simulation implementation method of confrontation network is generated based on condition |
CN110443867A (en) * | 2019-08-01 | 2019-11-12 | 太原科技大学 | Based on the CT image super-resolution reconstructing method for generating confrontation network |
CN110569796A (en) * | 2019-09-09 | 2019-12-13 | 南京东控智能交通研究院有限公司 | Method for dynamically detecting lane line and fitting lane boundary |
CN111899172A (en) * | 2020-07-16 | 2020-11-06 | 武汉大学 | Vehicle target detection method oriented to remote sensing application scene |
Non-Patent Citations (3)
Title |
---|
张宏钊: "基于加权损失函数的多尺度对抗网络图像语义分割算法", 《计算机应用与软件》 * |
蒋文杰: "基于对抗网络遥感图像超分辨率重建研究", 《计算机工程与应用》 * |
许辉: "基于高分影像与深度学习方法的路网提取技术研究与应用", 《公路》 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN118052869A (en) * | 2024-04-15 | 2024-05-17 | 深圳市峰和数智科技有限公司 | Unmanned aerial vehicle pose parameter optimization method and device, storage medium and computer equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112287807B (en) | Remote sensing image road extraction method based on multi-branch pyramid neural network | |
CN110263705B (en) | Two-stage high-resolution remote sensing image change detection system oriented to remote sensing technical field | |
CN109493320B (en) | Remote sensing image road extraction method and system based on deep learning, storage medium and electronic equipment | |
CN111598823A (en) | Multi-source mobile measurement point cloud data air-ground integrated fusion method and storage medium | |
CN111104850B (en) | Remote sensing image building automatic extraction method and system based on residual error network | |
CN111310756A (en) | Damaged corn particle detection and classification method based on deep learning | |
CN113160184B (en) | Unmanned aerial vehicle intelligent inspection cable surface defect detection method based on deep learning | |
CN104809724A (en) | Automatic precise registration method for multiband remote sensing images | |
CN111008641B (en) | Power transmission line tower external force damage detection method based on convolutional neural network | |
CN113554595A (en) | Unmanned aerial vehicle laser radar point cloud tower head deformation detection device and method | |
Wang et al. | Insulator defect recognition based on faster R-CNN | |
CN113838064A (en) | Cloud removing method using multi-temporal remote sensing data based on branch GAN | |
CN114283137A (en) | Photovoltaic module hot spot defect detection method based on multi-scale characteristic diagram inference network | |
CN113313107A (en) | Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge | |
CN115641504A (en) | Automatic remote sensing extraction method for field boundary based on crop phenological characteristics and decision tree model | |
CN112906459A (en) | Road network checking technology based on high-resolution remote sensing image and deep learning method | |
Qin et al. | Deep learning for filtering the ground from ALS point clouds: A dataset, evaluations and issues | |
CN111667461A (en) | Method for detecting abnormal target of power transmission line | |
CN114387261A (en) | Automatic detection method suitable for railway steel bridge bolt diseases | |
CN116664823A (en) | Small sample SAR target detection and recognition method based on meta learning and metric learning | |
CN116310802A (en) | Method and device for monitoring change of residence based on multi-scale fusion model | |
Cui et al. | Joint Superpixel Segmentation and Graph Convolutional Network Road Extration for High-Resolution Remote Sensing Imagery | |
CN114596490A (en) | Hilly land feature line extraction method and hilly land DEM (digital elevation model) fine production method | |
CN116189008A (en) | Remote sensing image change detection method based on fixed point number quantification | |
CN113609913B (en) | Pine wood nematode disease tree detection method based on sampling threshold interval weighting |
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 |