CN111242231A - Strip mine road model construction method based on P-LinkNet network - Google Patents
Strip mine road model construction method based on P-LinkNet network Download PDFInfo
- Publication number
- CN111242231A CN111242231A CN202010054020.2A CN202010054020A CN111242231A CN 111242231 A CN111242231 A CN 111242231A CN 202010054020 A CN202010054020 A CN 202010054020A CN 111242231 A CN111242231 A CN 111242231A
- Authority
- CN
- China
- Prior art keywords
- mine
- images
- model
- image
- network
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
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
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Abstract
A method for constructing a surface mine road model based on a P-LinkNet network includes acquiring aerial images of surface mines under different types of conditions, finding out characteristics of mine roads, updating existing training sets according to specific mine conditions, extracting paths by using the P-LinkNet network-based mine road model construction method, adding cavity convolution between encoding and decoding to increase perception fields of central features of the images, improving recognition rate of the images by adopting a cycle structure formed by adjusting the training sets, performing five-time cross validation on the models in a training process, training out a deep neural network model with high fault tolerance, and extracting paths of different mine pictures by using migration learning. And in consideration of the variability of the mine road, the mine road is modeled through the mine path extracted from the mine image, and the requirement for scheduling and optimizing the mine is met.
Description
Technical Field
The invention belongs to the technical field of artificial intelligence, relates to road extraction of image segmentation in deep learning, and particularly relates to a P-LinkNet network-based open-pit mine road model construction method.
Background
In recent years, the construction of smart mines is actively developed, the scheduling optimization of vehicles plays an important role in the construction of smart mines, and mine path extraction is an obstacle to scheduling.
The current path extraction technology mainly aims at urban roads and rural roads, and the two roads are greatly different from mine roads. Most urban roads are structured roads, although the layout is complex, the identification is simple; the general roads in rural areas are fewer, and the identification process is relatively easy; the mine road not only has a complex layout structure, but also is mostly driven by trucks, and the extraction of the mine route is more complex because the road is made of local ore waste.
Along with the change of the mining progress of the mine, the temporary road of the mine has great change, and a permanent road model cannot be established once and for all, so the mine road model needs to be updated regularly. At present, a mine road model is simply built mainly through a Google map and a manual measurement means, a large amount of manpower and material resources are consumed, and the method cannot be used for accurately modeling the mine road well.
How to solve the problem that the mine road model is regularly updated and built accurately only in the mine construction background is difficult even if the mine road model is hot in the scheduling optimization process.
Disclosure of Invention
In order to overcome the defects of the prior art and solve the problem of accurately building a mine road model, the invention aims to provide a P-LinkNet network-based strip mine road model building method, which aims to extract the path of a strip mine by means of a strip mine image.
In order to achieve the purpose, the invention adopts the technical scheme that:
a method for constructing an open pit mine road model based on a P-LinkNet network comprises the following steps:
step 1) acquiring aerial images of surface mines under different types of conditions, finding out characteristics of mine roads in the aerial images, and updating an existing training set according to specific mine conditions;
step 2) adjusting the composition of different types of images in the training set, performing feedback training on the P-LinkNet network, and verifying the network performance by adopting five-fold cross validation to obtain a trained P-LinkNet network model;
and 3) taking the selected latest mine image as a test set, testing the test set by using the P-LinkNet trained in the step 2), outputting an extracted path, and building a model of the whole mine according to the spatial position information of the image.
The step 1) specifically comprises the following steps:
step 1.1) acquiring images in the same area by using an unmanned aerial vehicle according to different illumination, different image shooting angles and different mine conditions;
step 1.2) dividing the shot images into a yin-yang two-class image according to the positive light and the negative light, wherein each class is divided into 6 groups, namely an ore permanent road, an ore temporary road, an earth permanent road, an earth temporary road, a non-road rock area and a non-road earth area, and each group is combined into four groups according to the shooting angle and mine field conditions, namely each group is divided into 4 subgroups again, and 2 × 6 × 4 subgroups are counted;
and step 1.3) adjusting a training set according to the model recognition rate and the cycle number returned after the network is trained so as to improve the recognition rate of the model.
In the step 1.3), a control variable method is adopted to adjust the training set, and a control variable method is adopted, wherein it is known that each class is 24 groups of 4 × 6, and 4 groups are one group, that is, 4 times of circulation can find out the optimal composition of one group.
The step 2) specifically comprises the following steps:
step 2.1) building two models which are sunny or sunny and cloudy or backlight; (since it is possible to mix the acquisitions only during the acquisition of the image, i.e. during one acquisition, with direct or backlit light, two models are constructed here)
Step 2.2) adding a hole convolution between LinkNet network coding and decoding to build a network;
step 2.3) carrying out secondary classification on the images through a sigmoid activation function, wherein the output foreground color is white, and the background color is black;
and 2.4) finding the optimal training set composition of a specific mine model by adjusting the training sets composed of different types of images, iterating for a certain number of times, finding the optimal training set, and obtaining the trained LinkNet network model capable of extracting paths. And a cycle consisting of changes formed by 24 training sets can be used for finding out the network with the best mine map identification efficiency through comparison.
In the invention, the image adopts a standardized 512-by-512 format, and the coding structure adopts ResNet 101.
After the image is coded, the image output is 16 × 16, three layers of cavity convolution are added to increase the receptive field of the central features of the image, and the expansion rates are 1, 2 and 4 respectively.
And 2) verifying the network performance by adopting five-fold cross validation in the step 2) to obtain a trained P-LinkNet network model, finding out images formed by a large number of optimal training sets based on the obtained optimal training sets, dividing the images into 5 groups according to requirements, performing five-time test set transformation by taking one group as a test set and taking the other 4 groups as training sets, and averaging the recognition rate of 5 times of tests to obtain the final recognition rate.
The selected latest mine image is used as a test set, the P-LinkNet network trained in the step 2) is used for testing the test set, and the output and extraction path specifically comprises the following steps: the method comprises the steps of shooting a specific mine on the spot, inputting shot images into a trained P-LinkNet network model, identifying related similar mine images by using the trained model through transfer learning, integrating all images identified in one mine, and outputting images extracted through a final path.
The building of the model of the whole mine according to the spatial position information of the image specifically comprises the following steps: and matching the original spatial position information of the image with the extracted image, and generating an image model of the whole mine by using map software.
Compared with the prior art, the method greatly saves manpower and material resources, and more efficiently and quickly constructs the mine road model.
Drawings
FIG. 1 is a flow chart of constructing a P-LinkNet network model according to the present invention.
Fig. 2 is a flow chart of the update of the training set.
FIG. 3 is a block diagram of the architecture of a P-LinkNet network.
Fig. 4 is a flowchart of building a mine road model.
Detailed Description
The embodiments of the present invention will be described in detail below with reference to the drawings and examples.
As shown in fig. 1 and 4, the method for constructing the strip mine road model based on the P-LinkNet network comprises the following steps:
step 1) acquiring aerial images of surface mines under different types of conditions, finding out characteristics of mine roads, and updating an existing training set according to specific mine conditions. The method comprises the following specific steps:
step 1.1) carrying out image acquisition in the same area by using an unmanned aerial vehicle according to different illumination, different image shooting angles and different mine conditions, and carrying out specific research on a mine to be modeled; examples are 1 normal light (sun-illuminated condition), 90 ° (angle of unmanned aerial vehicle or satellite photography), Henan molybdenum ore (mine site condition), 2 normal light, 90 °, Inmunoge, 3 normal light, 45 ° (angle of unmanned aerial vehicle or satellite photography), Henan molybdenum ore, 4 normal light, 45 °, Inmunoge, 5 backlight (shaded or cloudy day collection condition), 90 °, Henan molybdenum ore, 6 backlight, 90 °, Inmunoge, 7 backlight, 45 °, Henan molybdenum ore, 8 backlight, 45 °, Inmunoge.
Step 1.2) the shot images are divided into two main categories of yin and yang according to the positive light and the negative light, each category is divided into 6 main groups, namely an ore permanent road, an ore temporary road, an earth permanent road, an earth temporary road, a non-road rock area and a non-road earth area, four categories are combined in each group according to the shooting angle and mine field conditions, namely, each main group is divided into 4 subgroups again, and 2 × 6 × 4 subgroups are counted in total, and the steps are as follows: an ore permanent road (angle 1, condition 1; angle 1, condition 2; angle 2, condition 1), an ore temporary road (angle 1, condition 1; angle 1, condition 2; angle 2, condition 1), an earth permanent road (angle 1, condition 1; angle 1, condition 2; angle 2, condition 1), an earth temporary road (angle 1, condition 1; angle 1, condition 2; angle 2, condition 1), a non-road rock region (angle 1, condition 1; angle 1, condition 2; angle 2, condition 1; angle 2, condition 1), a non-road earth region (angle 1, condition 1; angle 1, condition 2; angle 2, condition 1).
And step 1.3) adjusting a training set according to the model recognition rate and the cycle number returned after the network is trained so as to improve the recognition rate of the model. Specifically, the training set is adjusted by a controlled variable method, wherein each class is known to be composed of 24 groups of 4 × 6, and 4 groups are known to be a large group, that is, the optimal composition of the large group can be found by cycling for 4 times. Specifically, the first input training set consists of:
(y1,y2,y3,……,y24＝1,0,0,0,1,0,0,0,1,0,0,0,1,0,0,0)
in the training set composition, y ═ 1 represents selection, and y ═ 0 represents non-selection. The max { rec1, rec2, rec3, rec4} (rec is the recognition rate of the model) is obtained four times per cycle, namely the model with the best recognition rate in the four models. The first loop with the highest recognition rate among the four times of y1 ', y2 ', y3 ', y4 ' is reserved, namely training set selection (y1 ', y2, y3 ', y4 ') -1, 0,0, and y1, y2, y3 and y4 after replacement. The following iteration can refer to table 1 specifically, and the 25 th loop is input to form an optimal training set. After circulation is carried out for 24 times, y1 ', y2 ', y3 and … … y24 ' can be found out; and in the 25 th circulation, the optimal training set can be obtained. Considering that the sequence of 6 groups of iterations can affect the final result, the sequence of 10 times of 6 groups of inputs is randomly generated, and finally the training set with the highest recognition rate is selected as the optimal training set, and fig. 2 is a flow chart of four cycles.
TABLE 1 example diagrams of training set one training set adjustment per time
Step 2) adjusting the composition of different types of images in the training set, and performing feedback training on the P-LinkNet network, wherein the method specifically comprises the following steps:
2.1) the photos shot under different weather conditions have great influence on mine images under different angles and different conditions, the colors displayed on the images have great difference, and the images can only be acquired in a mixed manner in the acquisition process, namely, in the primary acquisition process, light rays are directly emitted or backlight, so that two models are constructed, namely a female model and a male model;
step 2.2) the mine road is relatively complex, and an optimized LinkNet is adopted to build a network, namely, a void convolution is added between LinkNet network coding and decoding, and an image standardized to 512 by 512 is taken as an example, and a coding structure adopts ResNet 101; the mine road has the natural characteristics of narrowness, connectivity, complexity, large span and the like, in order to prevent the loss of information in the process of image pooling and deconvolution, as the input image is 512 × 512, after encoding, the image output is 16 × 16, three layers of cavity convolution are added to increase the receptive field of the central feature of the image, and the expansion rates are respectively 1, 2 and 4.
Step 2.3) carrying out secondary classification on the images through a sigmoid activation function, wherein the output foreground color is white, and the background color is black;
and 2.4) finding the optimal training set composition of the specific mine model by adjusting the training sets composed of different types of images, iterating for a certain number of times, finding the optimal training set, and further obtaining the optimal path extraction P-LinkNet network model.
And then verifying the network performance by adopting five-fold cross validation to obtain a trained P-LinkNet network model, finding out images formed by a large number of optimal training sets based on the obtained optimal training sets, dividing the images into 5 groups according to requirements, taking one group as a test set, taking the other 4 groups as training sets, carrying out five times of test set transformation, and averaging the recognition rate of 5 times of tests to obtain the final recognition rate.
And 3) taking the selected latest mine image as a test set, testing the test set by using the P-LinkNet network trained in the step 2), outputting an extracted path, and building a model of the whole mine according to the spatial position information of the image through corresponding software. The method specifically comprises the following steps:
the method comprises the steps of shooting a specific mine on the spot, inputting shot images into a trained PLinkNet network model, identifying related similar mine images by using the trained model through transfer learning, integrating all images identified in one mine, and outputting images extracted by a final path. And matching the original spatial position information of the image with the extracted image, and generating an image model of the whole mine by using map software.
In an embodiment of the present invention, an opencut road model is constructed by taking a valency molybdenum ore in the south of the river as an example (the model is constructed by adding an inner Mongolia gold ore as a comparison condition in order to eliminate interference of other conditions and increase redundancy of a mine model), and the specific steps are as follows:
1. and collecting different illumination, different image shooting angles and different mine conditions in the same area by using the unmanned aerial vehicle to acquire images. For the collected results, the image is preliminarily selected by using the existing P-LinkNet network, and the optimal training set is finally output as shown in the table 1 (the optimal training set of the positive model is shown)
2. Creating a training set with the optimal P-LinkNet network recognition rate in the step 1 as far as possible, and collecting images of mines and similar mines needing to be constructed in a large quantity.
3. The analysis of the directional light and the backlight has a great influence on the discrimination of mine roads. The collected images were segmented into 512 x 512 and then classified according to different lighting, different image capture angles and different mine conditions. And finally, constructing directional light and backlight training sets respectively for training a yin-yang network model and a yang network model.
4. Permanent roads are more than temporary roads in quantity, the permanent roads in mining areas are easy to distinguish, the temporary roads are not easy to distinguish, and the temporary roads are also the difficulty of path extraction, so that the overall composition of the roads in the training set is as follows: permanent/temporary road 1/2.
5. The composition of the ore and the soil of the mine is similar to the real composition of the specific mine.
6. The key point of the invention is to construct a road, so the set parameters are as follows: a road/non-road 5/1 training set was constructed according to the above configuration.
7. And on the basis of the 3-6 steps, randomly forming 5 groups of training sets according to the formation requirements according to mine images acquired according to the formation proportion of the final training set for subsequent 5-fold cross validation.
8. A P-LinkNet network is constructed, a specific network frame is shown in the attached drawing 3 in detail, and the specific steps are as follows:
8.1, the decoding module uses ResNet101 to set the size of the input image to 512 x 512. The following table specifically shows:
8.2, after the input image passes through the encoder, 2048 feature modules with the size of 16 × 16 are output
And 8.3, considering the natural characteristics of connectivity and the like of the road, adding a hole convolution layer to prevent information loss of the road characteristics in the processes of pooling and deconvolution, increasing the receptive field of the characteristic image and better keeping the characteristics of the road under the condition of ensuring that the size of the characteristic image is not changed.
And 8.4, because the size of the output of the encoder is 16 x 16, three layers of hole convolution are added, and a parallel mode of cascade connection and series connection is adopted.
8.5, decoding the characteristics by using a LinkNet network, and finally outputting 256 × 64
8.6 changing the image to 512 x 32 by deconvolution
8.7, finally selecting an activation function as sigmoid to carry out secondary classification on the image, and outputting the extracted path
9. And after 24 times of iteration, finding out an optimal training set composition, reconstructing the neural network, finally selecting an optimal model, and outputting a yin-yang model.
10. And (3) forming the acquired new mine images according to the optimal training set, dividing the acquired new mine images into 5 groups, randomly removing 1 group from the test set, inputting the training set with 4 groups into a P-linknet network, performing 5-time test, completing 5-fold cross validation, and referring to the accuracy of a subsequent model.
11. For the input of different kinds of images, the extraction effect of the images after the path extraction is not different, so the images output by the two final models are merged together.
12. And combining the image after the path is extracted with the original image space information.
13. And based on the combined images, performing image processing by using combined map software, and finally outputting the mine model.
Claims (9)
1. A P-LinkNet network-based strip mine road model construction method is characterized by comprising the following steps:
step 1) acquiring aerial images of surface mines under different types of conditions, finding out characteristics of mine roads in the aerial images, and updating an existing training set according to specific mine conditions;
step 2) adjusting the composition of different types of images in the training set, performing feedback training on the P-LinkNet network, and verifying the network performance by adopting five-fold cross validation to obtain a trained P-LinkNet network model;
and 3) taking the selected latest mine image as a test set, testing the test set by using the P-LinkNet trained in the step 2), outputting an extracted path, and building a model of the whole mine according to the spatial position information of the image.
2. The P-LinkNet network-based strip mine road model building method according to claim 1, wherein the step 1) specifically comprises:
step 1.1) acquiring images in the same area by using an unmanned aerial vehicle according to different illumination, different image shooting angles and different mine conditions;
step 1.2) dividing the shot images into a yin-yang two-class image according to the positive light and the negative light, wherein each class is divided into 6 groups, namely an ore permanent road, an ore temporary road, an earth permanent road, an earth temporary road, a non-road rock area and a non-road earth area, and each group is combined into four groups according to the shooting angle and mine field conditions, namely each group is divided into 4 subgroups again, and 2 × 6 × 4 subgroups are counted;
and step 1.3) adjusting a training set according to the model recognition rate and the cycle number returned after the network is trained so as to improve the recognition rate of the model.
3. The method for constructing the open-pit mine road model based on the P-LinkNet network according to claim 2, wherein in the step 1.3), a control variable method is adopted to adjust the training set, and a control variable method is adopted, wherein each type is known to be 24 groups of 4 by 6, and 4 groups are a large group, namely, 4 times of circulation can find out the optimal composition of the large group.
4. The P-LinkNet network-based strip mine road model building method according to claim 1, wherein the step 2) specifically comprises:
step 2.1) building two models which are sunny or sunny and cloudy or backlight; (since it is possible to mix the acquisitions only during the acquisition of the image, i.e. during one acquisition, with direct or backlit light, two models are constructed here)
Step 2.2) adding a hole convolution between LinkNet network coding and decoding to build a network;
step 2.3) carrying out secondary classification on the images through a sigmoid activation function, wherein the output foreground color is white, and the background color is black;
and 2.4) finding the optimal training set composition of a specific mine model by adjusting the training sets composed of different types of images, iterating for a certain number of times, finding the optimal training set, and obtaining the trained LinkNet network model capable of extracting paths.
5. The P-LinkNet network-based strip mine road model construction method according to claim 4, wherein the images are in a standardized 512-by-512 format, and the coding structure is ResNet 101.
6. The P-LinkNet network-based strip mine road model construction method according to claim 5, wherein the image is encoded, the image output is 16 x 16, three layers of hole convolution are added to increase the receptive field of the central features of the image, and the expansion rates are 1, 2 and 4 respectively.
7. The method for constructing the open pit mine road model based on the P-LinkNet network according to claim 5, wherein five-fold cross validation is adopted in the step 2) to verify the network performance, a trained P-LinkNet network model is obtained, based on the obtained optimal training set, images formed by a large number of optimal training sets are found, the images are divided into 5 groups according to requirements, one group is used as a test set, the other 4 groups are used as training sets, five times of test set transformation is carried out, and the recognition rate of 5 times of tests is averaged to be used as the final recognition rate.
8. The method for constructing the strip mine road model based on the P-LinkNet network according to claim 1, wherein the step of testing the test set by using the P-LinkNet network trained in the step 2) by using the selected latest mine image as the test set and outputting the extracted path specifically comprises the steps of: the method comprises the steps of shooting a specific mine on the spot, inputting shot images into a trained P-LinkNet network model, identifying related similar mine images by using the trained model through transfer learning, integrating all images identified in one mine, and outputting images extracted through a final path.
9. The P-LinkNet network-based strip mine road model building method according to claim 1 or 7, wherein the building of the model of the whole mine according to the spatial position information of the image specifically comprises: and matching the original spatial position information of the image with the extracted image, and generating an image model of the whole mine by using map software.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010054020.2A CN111242231B (en) | 2020-01-17 | 2020-01-17 | Strip mine road model construction method based on P-LinkNet network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010054020.2A CN111242231B (en) | 2020-01-17 | 2020-01-17 | Strip mine road model construction method based on P-LinkNet network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111242231A true CN111242231A (en) | 2020-06-05 |
| CN111242231B CN111242231B (en) | 2023-05-23 |
Family
ID=70872755
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010054020.2A Active CN111242231B (en) | 2020-01-17 | 2020-01-17 | Strip mine road model construction method based on P-LinkNet network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111242231B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112347976A (en) * | 2020-11-23 | 2021-02-09 | 腾讯科技（深圳）有限公司 | Region extraction method and device for remote sensing satellite image, electronic equipment and medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050147292A1 (en) * | 2000-03-27 | 2005-07-07 | Microsoft Corporation | Pose-invariant face recognition system and process |
| US20180286038A1 (en) * | 2015-09-23 | 2018-10-04 | The Regents Of The University Of California | Deep learning in label-free cell classification and machine vision extraction of particles |
| CN109446970A (en) * | 2018-10-24 | 2019-03-08 | 西南交通大学 | A kind of Intelligent Mobile Robot road scene recognition methods based on deep learning |
-
2020
- 2020-01-17 CN CN202010054020.2A patent/CN111242231B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20050147292A1 (en) * | 2000-03-27 | 2005-07-07 | Microsoft Corporation | Pose-invariant face recognition system and process |
| US20180286038A1 (en) * | 2015-09-23 | 2018-10-04 | The Regents Of The University Of California | Deep learning in label-free cell classification and machine vision extraction of particles |
| CN109446970A (en) * | 2018-10-24 | 2019-03-08 | 西南交通大学 | A kind of Intelligent Mobile Robot road scene recognition methods based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 李云伍等: "基于改进空洞卷积神经网络的丘陵山区田间道路场景识别", 《农业工程学报》 * |
| 陆永帅等: "深度置信网络模型的机载多光谱数据罂粟识别", 《遥感信息》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112347976A (en) * | 2020-11-23 | 2021-02-09 | 腾讯科技（深圳）有限公司 | Region extraction method and device for remote sensing satellite image, electronic equipment and medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111242231B (en) | 2023-05-23 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110619282B (en) | Automatic extraction method for unmanned aerial vehicle orthoscopic image building | |
| CN106778527B (en) | A kind of improved neural network pedestrian recognition methods again based on triple losses | |
| CN111259906B (en) | Method for generating remote sensing image target segmentation countermeasures under condition containing multilevel channel attention | |
| CN109063723A (en) | The Weakly supervised image, semantic dividing method of object common trait is excavated based on iteration | |
| CN109614968B (en) | License plate detection scene picture generation method based on multi-scale mixed image stylization | |
| CN109635748B (en) | Method for extracting road characteristics in high-resolution image | |
| CN105930868A (en) | Low-resolution airport target detection method based on hierarchical reinforcement learning | |
| CN113902915A (en) | Semantic segmentation method and system based on low-illumination complex road scene | |
| CN111832615A (en) | Sample expansion method and system based on foreground and background feature fusion | |
| CN108647682A (en) | A kind of brand Logo detections and recognition methods based on region convolutional neural networks model | |
| CN107194349A (en) | High-resolution PolSAR images city terrain classification method based on deep learning | |
| CN115527123B (en) | Land cover remote sensing monitoring method based on multisource feature fusion | |
| CN110378047A (en) | A kind of Longspan Bridge topology ambiguity three-dimensional rebuilding method based on computer vision | |
| Chen et al. | ASF-Net: Adaptive screening feature network for building footprint extraction from remote-sensing images | |
| Peng et al. | A proxy model to predict reservoir dynamic pressure profile of fracture network based on deep convolutional generative adversarial networks (DCGAN) | |
| CN114463340B (en) | Agile remote sensing image semantic segmentation method guided by edge information | |
| CN111242231B (en) | Strip mine road model construction method based on P-LinkNet network | |
| CN113963104B (en) | Three-dimensional model generation method of layout image generated based on GAN model | |
| CN104537660B (en) | Image partition method based on Multiobjective Intelligent body evolution clustering algorithm | |
| Jiang et al. | AGD-Linknet: a road semantic segmentation model for high resolution remote sensing images integrating attention mechanism, gated decoding block and dilated convolution | |
| CN113159044A (en) | Deep learning-based road material identification method for convolutional neural network | |
| CN114626294A (en) | Planning layout hierarchical generation method for university campus | |
| CN114565730A (en) | Method and device for processing map vector sample and computer readable storage medium | |
| CN110738249A (en) | aurora image clustering method based on deep neural network | |
| CN117315614B (en) | Traffic target detection method based on improved YOLOv7 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |