CN111209894A - Roadside illegal building identification method for road aerial image - Google Patents
Roadside illegal building identification method for road aerial image Download PDFInfo
- Publication number
- CN111209894A CN111209894A CN202010084213.2A CN202010084213A CN111209894A CN 111209894 A CN111209894 A CN 111209894A CN 202010084213 A CN202010084213 A CN 202010084213A CN 111209894 A CN111209894 A CN 111209894A
- Authority
- CN
- China
- Prior art keywords
- road
- suspected
- building
- detected
- target
- 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
- 238000000034 method Methods 0.000 title claims abstract description 27
- 238000003062 neural network model Methods 0.000 claims abstract description 20
- 238000012549 training Methods 0.000 claims abstract description 6
- 230000011218 segmentation Effects 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 4
- 238000012216 screening Methods 0.000 claims description 3
- 101000615238 Homo sapiens Proto-oncogene DBL Proteins 0.000 description 15
- 238000013527 convolutional neural network Methods 0.000 description 14
- 239000013598 vector Substances 0.000 description 12
- 238000011176 pooling Methods 0.000 description 10
- 238000001514 detection method Methods 0.000 description 9
- 238000013528 artificial neural network Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 3
- 101100194362 Schizosaccharomyces pombe (strain 972 / ATCC 24843) res1 gene Proteins 0.000 description 2
- 101100194363 Schizosaccharomyces pombe (strain 972 / ATCC 24843) res2 gene Proteins 0.000 description 2
- 238000012790 confirmation Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000004927 fusion Effects 0.000 description 2
- 239000013589 supplement Substances 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000002789 length control Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000011084 recovery Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
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
- G06V20/176—Urban or other man-made structures
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A roadside illegal building identification method of a road aerial image relates to the technical field of road administration facilities and solves the technical problem of improving identification efficiency. The method comprises the steps of constructing a road picture sample library and a suspected violation picture sample library of a violation building by utilizing aerial pictures of all road sections of a target road; and training the deep neural network model by using the constructed sample library, and identifying and segmenting the aerial image of the to-be-detected road section of the target road by using the trained deep neural network model to identify the illegal building. The identification method provided by the invention is used for identifying the illegal buildings on two sides of the road, and can improve the identification efficiency of the illegal buildings.
Description
Technical Field
The invention relates to a road administration facility technology, in particular to a roadside illegal building identification method technology of a road aerial image.
Background
The road length system is a city management mode, which effectively promotes city management to be refined according to the principles of one-road-length, one-block combination and property management, and under the road length system management mode, in order to avoid citizens or organizations building illegal buildings, illegal buildings on two sides of a road need to be identified and the approximate positions of the illegal buildings need to be determined so as to facilitate on-site confirmation and supervision of patrolling personnel.
The traditional monitoring and management of illegal buildings on two sides of a road mainly depends on the report, discovery and confirmation of the public, and the illegal buildings are forbidden and cannot be discovered in time, so that the development and construction of cities are hindered.
At present, some methods identify the roadside illegal buildings by analyzing road images, but the existing identification methods compare the acquired images with the original historical images one by one, the historical images need to be repeatedly called in the identification method, most of the historical images need to be identified and detected by combining manpower, resources are consumed, and the efficiency is reduced.
Disclosure of Invention
Aiming at the defects in the prior art, the invention aims to provide a roadside illegal building identification method for a road aerial image with high identification efficiency.
In order to solve the technical problem, the invention provides a roadside illegal building identification method of a road aerial image, which is characterized by comprising the following specific steps of:
s1: acquiring aerial pictures of each section of a target road in an aerial shooting mode to obtain a picture sample library of the target road;
s2: manually screening road pictures of all road sections of the target road and pictures of suspected illegal buildings from a picture sample library of the target road, forming a road picture sample library by using the screened road pictures, and forming a illegal picture sample library by using the screened pictures of the suspected illegal buildings;
s3: constructing a deep neural network model, and training the constructed deep neural network model by respectively taking each picture in the road picture sample library and the violation picture sample library as an input image until the deep neural network model is trained to a preset target parameter;
s4: acquiring an aerial photo of the to-be-detected section of the target road in an aerial photographing mode, and identifying and segmenting the aerial photo of the to-be-detected section of the target road by using a trained deep neural network model;
if suspected illegal buildings exist in the aerial image of the to-be-detected road section of the target road, the deep neural network model outputs the detected position area of each suspected illegal building and the road segmentation mask of the to-be-detected road section of the target road; calculating the area of each suspected violation building and the distance between each suspected violation building and the road section to be detected of the target road according to the position area where each suspected violation building is located and the road segmentation mask of the road section to be detected of the target road;
and presetting a distance discrimination threshold, and if the distance calculation value between the suspected violation building and the road section to be detected of the target road is smaller than the preset distance discrimination threshold for each suspected violation building, judging the suspected violation building as the violation building.
According to the roadside illegal building identification method of the road aerial image, the screened aerial sample picture is used for training to obtain the network model with relatively reasonable parameter weight, and the network model is used for identifying other roads to be detected and suspected illegal buildings, so that the accuracy and the detection efficiency are greatly improved, meanwhile, manual detection is avoided, manpower and material resources are saved, the method is high in adaptability, can be well adapted to changes of external environments such as weather, and has the advantages of being good in stability, high in operation efficiency, accurate in detection, capable of saving manpower and strong in robustness.
Drawings
Fig. 1 is a flow chart of the identification of a roadside violation building identification method of a road aerial image according to an embodiment of the invention.
Detailed Description
The following description will be made in detail with reference to the accompanying drawings, but the present invention is not limited thereto, and all similar structures and similar variations thereof adopted by the present invention shall fall within the protection scope of the present invention, wherein the pause numbers in the present invention shall represent the relation of the pause numbers, and the english letters in the present invention shall be distinguished by the case.
As shown in fig. 1, the roadside violation building identification method of the road aerial image provided by the embodiment of the invention is characterized by comprising the following specific steps:
s1: acquiring aerial pictures of all sections of a target road in an aerial shooting mode to obtain a picture sample library of the target road, wherein the aerial shooting target road is preferably clear and cloudy, and overcast and rainy weather is avoided as much as possible;
s2: manually screening road pictures of all road sections of the target road and pictures of suspected illegal buildings from a picture sample library of the target road, forming a road picture sample library by using the screened road pictures, and forming a illegal picture sample library by using the screened pictures of the suspected illegal buildings;
s3: constructing a deep neural network model, and training the constructed deep neural network model by respectively taking each picture in the road picture sample library and the violation picture sample library as an input image until the deep neural network model is trained to a preset target parameter;
the deep neural network model and the method for training the deep neural network model are both the prior art, the deep neural network is a discrimination model, is provided with at least one hidden layer of neural network, and can be trained by using a back propagation algorithm;
s4: acquiring an aerial photo of the to-be-detected section of the target road in an aerial photographing mode, and identifying and segmenting the aerial photo of the to-be-detected section of the target road by using a trained deep neural network model;
if suspected illegal buildings exist in the aerial image of the to-be-detected road section of the target road, the deep neural network model outputs the detected position area of each suspected illegal building and the road segmentation mask of the to-be-detected road section of the target road; calculating the area of each suspected violation building and the distance between each suspected violation building and the road section to be detected of the target road according to the position area where each suspected violation building is located and the road segmentation mask of the road section to be detected of the target road;
and presetting a distance discrimination threshold, and if the distance calculation value between the suspected violation building and the road section to be detected of the target road is smaller than the preset distance discrimination threshold for each suspected violation building, judging the suspected violation building as the violation building.
The deep neural network model constructed in the step S3 of the embodiment of the invention comprises three modules, wherein the three modules are a road segmentation module, a violation building detection module and a fusion ranging module respectively;
1) the road segmentation module adopts a full convolution network FCN, the structure of which is similar to a convolution neural network CNN structure, the main difference is that a full connection layer is cancelled and replaced by a convolution layer, a heat map is finally output instead of a feature map, and meanwhile, in order to solve the influence of convolution and pooling on the image size, an up-sampling mode is used for recovery, and the specific structure is as follows:
and (3) convolution: in the conventional classified network, a full connection layer is usually used at last, an original two-dimensional feature map is converted into a one-dimensional feature vector with a fixed length, so that spatial information is lost, and a vector with a specific length is output at last to represent the probability that an input image belongs to each class and serve as a classified label; unlike the conventional convolutional neural network CNN which uses a full link layer to obtain a feature vector of a fixed length for classification (full link layer + softmax) after a convolutional layer, the full convolutional network FCN can accept an input image of any size, then up-samples a feature map (feature map) of the last convolutional layer through a convolutional layer to restore the feature map to the same size as the input image, so that a prediction can be generated for each pixel, and spatial information in the original input image is retained.
The full convolutional network FCN converts the fully connected layers in the conventional convolutional neural network CNN into convolutional layers, corresponding to the convolutional neural network CNN, the full convolutional network FCN converts the last three fully connected layers into three convolutional layers, in the conventional convolutional neural network CNN structure, the first 5 layers are convolutional layers, the 6 th and 7 th layers are respectively one-dimensional vectors with length of 4096, the 8 th layer is one-dimensional vectors with length of 1000, respectively corresponding to 1000 different types of probabilities, the full convolutional network FCN converts the 3 layers into convolutional layers, the sizes of convolutional cores (number of channels, width, and height) are respectively (4096, 1, 1), (1000, 1, 1), and there is no difference in number, but convolutional and full connection is a different concept and calculation process, and uses weights and offsets that have been trained in the convolutional neural network CNN before, but not only that the weights and offsets have their own ranges, belonging to its own convolution kernel, so all layers in the full convolution network FCN are convolution layers, and are called full convolution network.
Deconvolution: the deconvolution process can be understood as an Upsampling process (Upsampling), a pooling layer of a conventional convolutional neural network CNN structure (such as AlexNet, VGGNet) reduces the size of a feature map, a segmented picture with the same size as an input image needs to be output in semantic segmentation, so that the feature map needs to be upsampled, and in order to obtain the feature map with the same size as the input image, a crop operation is also used in a full convolutional network to assist a deconvolution operation because the deconvolution operation does not exactly enlarge the feature map by an integral multiple.
A hopping structure: the original image is reduced to 1/2 after convolution conv1 (vector convolution operation) and pool1 (pooling) are carried out on the original image; then, the image is subjected to second conv2 (vector convolution operation) and pool2 (pooling), and the image is reduced to 1/4; continuing to perform a third convolution operation conv3 (vector convolution operation) and pool3 (pooling) on the image to reduce the image into 1/8 of the original image, and retaining a feature Map (feature Map) of the pool3 (pooling); continuing to perform a fourth convolution operation conv4 (vector convolution operation) and pool4 (pooling) on the image, reducing the image into 1/16 of the original image, and reserving a feature Map (feature Map) of the pool4 (pooling); finally, the fifth convolution operation conv5 (vector convolution operation) and pool5 (pooling) are performed on the image, the image is reduced to 1/32 of the original image, then the full connection in the original convolution neural network CNN operation is changed into convolution operations conv6 (vector convolution operation) and conv7 (vector convolution operation), the number of feature maps (feature maps) of the image is changed but the image size is still 1/32 of the original image, at this time, the image is not called feature Map (feature Map) any more, but called heat Map (heat Map), heat Map (heat Map) with 1/32 size, feature Map (feature Map) with 1/16 size and feature Map (feature Map) with 1/8 size, after the Upsampling process (Upsampling) operation is performed on the heat Map (heat Map) with 1/32 size, because the restored pictures by such operations are only the features in the conv5, the convolution accuracy problem cannot be well restored among the features in the image, therefore, in the previous iteration, the convolution kernel in conv4 is deconvolved to supplement details (equivalent to an interpolation process) on the graph after the last Upsampling process (Upsampling), and finally the convolution kernel in conv3 is deconvolved again to supplement details on the image after the last Upsampling process (Upsampling), so that the restoration of the whole image is completed, specifically, the results of different pooling layers are upsampled, and then the results are combined to optimize the output.
2) The violation building detection module adopts a full convolution network, wherein a plurality of residual jump layer connection modes are provided, and the specific structure comprises the following components:
feature extractor (Darknet-53): the module is characterized in that YOLOv2, Darknet-19 and a residual network structure are fused, the module is formed by combining continuous convolution layers of 3 × 3 and 1 × 1, a few shortcut connections are also added, the whole quantity is larger, the module is also called Darknet-53 because a total of 53 convolution layers, and as a main structure of the module, the Darknet-53 main structure sequentially comprises DBL, res1, res2, res8, res8 and res4, wherein the DBL comprises convolution layers, batch standardization and an activation function LeakyReLU, res1 comprises a zero filling layer, a DBL layer and 1 res unit layer, one res unit comprises two DBLs, res2 comprises a zero filling layer, a DBL layer and 2 res unit layers, res8 comprises a zero filling layer, a DBL layer and 8 res units, and res 24 res4 res layers comprise zero filling layers and 864 res layers;
res4 in Darknet-53 is output and connected with 5 DBLs, and after output, y1 is output after 1 DBL and 1 convolutional layer are obtained; simultaneously, tensor splicing is carried out on the output through 1 DBL and upsampling and the output of the second res8 in Darknet-53, and y2 is output after 5 DBLs, 1 DBLs and convolutional layers; meanwhile, the value after splicing and 5 DBLs is subjected to tensor splicing with the output of the 1 st res8 in Darknet-53 through 1 DBL and upsampling, and then the value after splicing is subjected to tensor splicing at the output y3 of 5 DBLs, 1 DBL and convolutional layer.
Cross-size prediction: the network provides 3 bounding boxes of different sizes, the system extracts features of these sizes with similar concepts to form a pyramid network, adds several convolutional layers in the basic feature extractor, and predicts a three-dimensional tensor code with the last convolutional layer: bounding boxes, in-box targets, and classification predictions. Then get the characteristic map from the first two layers, and sample it for 2 times, and get the characteristic map from the layer earlier in the network, link the characteristic maps of two kinds of resolutions of high and low together with element-wise, so can make us find the up-sampled characteristic and fine-grained characteristic in the early characteristic map, and obtain the more meaningful semantic information, later add several convolution layers to process this characteristic map combination, and finally predict a similar tensor whose size is twice of the original one, the clustering method that the network uses is K-Means, it can be used for confirming the priori of the bounding box.
The feature map (feature map) size of the module network output is y 1: (13 × 13), y 2: (26 × 26), y 3: (52 × 52), the network receives one (416 × 416) graph, down-samples by 5 convolutions with step size 2 (416/2 ˆ 5= 13, y1 outputs (13 × 13), up-samples from the convolutional layer of the second-to-last layer of y1 (x2, up sampling) are connected to the last feature map tensor of size 26 × 26, y2 outputs (26 × 26), up-samples from the convolutional layer of the second-to-last layer of y2 (x2, up sampling) are connected to the last feature map tensor of size 52 × 52, and y3 outputs (52 × 52).
3) Fuse the range finding module: the module is mainly used for designing a fusion algorithm according to results obtained by target detection and semantic segmentation, and the algorithm can jointly determine whether a detection result is a violation building according to the context, distance and target size in the target detection result, so that the accuracy of target detection is improved.
The embodiment of the invention provides a method for identifying illegal buildings on two sides of an aerial photography road under the road length control, which has the following main theoretical basis: the characteristics of the image are considered to comprise shallow pixels, deep linear structures such as straight lines and curves in various shapes, deep planar structures such as rectangles, triangles and circles, and aggregation of multiple planar structures, so that various texture characteristics are formed, namely, low-level characteristics are combined to form more abstract high-level characteristics, and finally, various high-level characteristics form various target categories which can be identified by human eyes and have practical significance.
Claims (1)
1. A roadside illegal building identification method of a road aerial image is characterized by comprising the following specific steps:
s1: acquiring aerial pictures of each section of a target road in an aerial shooting mode to obtain a picture sample library of the target road;
s2: manually screening road pictures of all road sections of the target road and pictures of suspected illegal buildings from a picture sample library of the target road, forming a road picture sample library by using the screened road pictures, and forming a illegal picture sample library by using the screened pictures of the suspected illegal buildings;
s3: constructing a deep neural network model, and training the constructed deep neural network model by respectively taking each picture in the road picture sample library and the violation picture sample library as an input image until the deep neural network model is trained to a preset target parameter;
s4: acquiring an aerial photo of the to-be-detected section of the target road in an aerial photographing mode, and identifying and segmenting the aerial photo of the to-be-detected section of the target road by using a trained deep neural network model;
if suspected illegal buildings exist in the aerial image of the to-be-detected road section of the target road, the deep neural network model outputs the detected position area of each suspected illegal building and the road segmentation mask of the to-be-detected road section of the target road; calculating the area of each suspected violation building and the distance between each suspected violation building and the road section to be detected of the target road according to the position area where each suspected violation building is located and the road segmentation mask of the road section to be detected of the target road;
and presetting a distance discrimination threshold, and if the distance calculation value between the suspected violation building and the road section to be detected of the target road is smaller than the preset distance discrimination threshold for each suspected violation building, judging the suspected violation building as the violation building.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010084213.2A CN111209894A (en) | 2020-02-10 | 2020-02-10 | Roadside illegal building identification method for road aerial image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010084213.2A CN111209894A (en) | 2020-02-10 | 2020-02-10 | Roadside illegal building identification method for road aerial image |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111209894A true CN111209894A (en) | 2020-05-29 |
Family
ID=70787489
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010084213.2A Pending CN111209894A (en) | 2020-02-10 | 2020-02-10 | Roadside illegal building identification method for road aerial image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111209894A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111931743A (en) * | 2020-10-09 | 2020-11-13 | 杭州科技职业技术学院 | Building violation monitoring method and system and electronic equipment |
| CN112215190A (en) * | 2020-10-21 | 2021-01-12 | 南京智慧航空研究院有限公司 | Illegal building detection method based on YOLOV4 model |
| CN112613437A (en) * | 2020-12-28 | 2021-04-06 | 国网浙江省电力有限公司电力科学研究院 | High-accuracy illegal building identification method |
| CN112651338A (en) * | 2020-12-26 | 2021-04-13 | 广东电网有限责任公司电力科学研究院 | Method and device for distinguishing hidden danger of illegal construction of power transmission line |
| CN114005038A (en) * | 2021-11-08 | 2022-02-01 | 浙江力石科技股份有限公司 | Method for identifying illegal buildings near scenic spot |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108710875A (en) * | 2018-09-11 | 2018-10-26 | 湖南鲲鹏智汇无人机技术有限公司 | A kind of take photo by plane road vehicle method of counting and device based on deep learning |
| CN108764137A (en) * | 2018-05-29 | 2018-11-06 | 福州大学 | Vehicle traveling lane localization method based on semantic segmentation |
| CN109426773A (en) * | 2017-08-24 | 2019-03-05 | 浙江宇视科技有限公司 | A kind of roads recognition method and device |
| CN110399820A (en) * | 2019-07-16 | 2019-11-01 | 西安工程大学 | A kind of margin of roads scenery visual identity analysis method |
-
2020
- 2020-02-10 CN CN202010084213.2A patent/CN111209894A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109426773A (en) * | 2017-08-24 | 2019-03-05 | 浙江宇视科技有限公司 | A kind of roads recognition method and device |
| CN108764137A (en) * | 2018-05-29 | 2018-11-06 | 福州大学 | Vehicle traveling lane localization method based on semantic segmentation |
| CN108710875A (en) * | 2018-09-11 | 2018-10-26 | 湖南鲲鹏智汇无人机技术有限公司 | A kind of take photo by plane road vehicle method of counting and device based on deep learning |
| CN110399820A (en) * | 2019-07-16 | 2019-11-01 | 西安工程大学 | A kind of margin of roads scenery visual identity analysis method |
Non-Patent Citations (2)
| Title |
|---|
| 袁延良: ""基于航空影像的城市违章建筑识别研究"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
| 陈一鸣: ""基于深度学习的遥感图像新增建筑物语义分割"", 《计算机与数字工程》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111931743A (en) * | 2020-10-09 | 2020-11-13 | 杭州科技职业技术学院 | Building violation monitoring method and system and electronic equipment |
| CN112215190A (en) * | 2020-10-21 | 2021-01-12 | 南京智慧航空研究院有限公司 | Illegal building detection method based on YOLOV4 model |
| CN112651338A (en) * | 2020-12-26 | 2021-04-13 | 广东电网有限责任公司电力科学研究院 | Method and device for distinguishing hidden danger of illegal construction of power transmission line |
| CN112651338B (en) * | 2020-12-26 | 2022-02-15 | 广东电网有限责任公司电力科学研究院 | Method and device for distinguishing hidden danger of illegal construction of power transmission line |
| CN112613437A (en) * | 2020-12-28 | 2021-04-06 | 国网浙江省电力有限公司电力科学研究院 | High-accuracy illegal building identification method |
| CN112613437B (en) * | 2020-12-28 | 2022-07-12 | 国网浙江省电力有限公司电力科学研究院 | Method for identifying illegal buildings |
| CN114005038A (en) * | 2021-11-08 | 2022-02-01 | 浙江力石科技股份有限公司 | Method for identifying illegal buildings near scenic spot |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111209894A (en) | Roadside illegal building identification method for road aerial image | |
| CN109934163B (en) | Aerial image vehicle detection method based on scene prior and feature re-fusion | |
| US20220222824A1 (en) | Fully automated multimodal system architecture for semantic segmentation of large-scale 3d outdoor point cloud data | |
| CN110163213B (en) | Remote sensing image segmentation method based on disparity map and multi-scale depth network model | |
| Delibaltov et al. | Parking lot occupancy determination from lamp-post camera images | |
| US9576373B2 (en) | Geospatial imaging system providing segmentation and classification features and related methods | |
| CN108416292B (en) | Unmanned aerial vehicle aerial image road extraction method based on deep learning | |
| CN112070070B (en) | LW-CNN method and system for urban remote sensing scene recognition | |
| CN113705636A (en) | Method and device for predicting trajectory of automatic driving vehicle and electronic equipment | |
| Narazaki et al. | Automated vision-based bridge component extraction using multiscale convolutional neural networks | |
| CN111144418A (en) | Railway track area segmentation and extraction method | |
| WO2020061648A1 (en) | Apparatus and method for three-dimensional object recognition | |
| CN112766409A (en) | Feature fusion method for remote sensing image target detection | |
| CN111369539A (en) | Building facade window detecting system based on multi-feature map fusion | |
| CN113486779A (en) | Panoramic intelligent inspection system for power transmission line | |
| CN115512222A (en) | Method for evaluating damage of ground objects in disaster scene of offline training and online learning | |
| CN114550023A (en) | Traffic target static information extraction device | |
| CN114898243A (en) | Traffic scene analysis method and device based on video stream | |
| CN113158954A (en) | Automatic traffic off-site zebra crossing area detection method based on AI technology | |
| CN114882490B (en) | Unlimited scene license plate detection and classification method based on point-guided positioning | |
| CN116129234A (en) | Attention-based 4D millimeter wave radar and vision fusion method | |
| CN112036246B (en) | Construction method of remote sensing image classification model, remote sensing image classification method and system | |
| CN114359705A (en) | Geological disaster monitoring method and device | |
| CN114639084A (en) | Road side end vehicle sensing method based on SSD (solid State disk) improved algorithm | |
| Alonso et al. | Automatic uav-based airport pavement inspection using mixed real and virtual scenarios |
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: 20200529 |
|
| RJ01 | Rejection of invention patent application after publication |