CN111461006B - Optical remote sensing image tower position detection method based on deep migration learning - Google Patents
Optical remote sensing image tower position detection method based on deep migration learning Download PDFInfo
- Publication number
- CN111461006B CN111461006B CN202010247180.9A CN202010247180A CN111461006B CN 111461006 B CN111461006 B CN 111461006B CN 202010247180 A CN202010247180 A CN 202010247180A CN 111461006 B CN111461006 B CN 111461006B
- Authority
- CN
- China
- Prior art keywords
- tower
- data
- remote sensing
- network
- source domain
- 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.)
- Active
Links
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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10032—Satellite or aerial image; Remote sensing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
-
- 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
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Molecular Biology (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Bioinformatics & Computational Biology (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
An optical remote sensing image tower position detection method based on deep migration learning belongs to the technical field of image processing, and comprises the following steps: step one: acquiring DOTA data and transmission line tower image data of a Pleiades satellite, and processing the acquired data according to a standard VOC data format; step two: designing a deep neural network structure for detecting towers, and providing a high-accuracy tower position detection neural network aiming at different tower states, background types, illumination and other scenes; step three: a deep migration learning method is introduced, so that the requirement of an algorithm on the target detection data of the remote sensing image tower is reduced, and model training is completed; step four: and testing and verifying the algorithm by using the remote sensing images containing the four tower states. The method is suitable for detecting the transmission tower with high accuracy in the visible light remote sensing image under complex application scenes such as various tower states, background types, illumination and the like, and provides support for safety monitoring of the transmission line.
Description
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to an optical remote sensing image tower position detection method based on deep migration learning.
The high-voltage transmission tower is one of the most important infrastructures in the power transmission system, the running state of the high-voltage transmission tower determines the running stability and safety of the whole power grid, and the detection of the high-voltage transmission tower is the basis of the running state monitoring of the high-voltage transmission tower and is also an important part of the monitoring of the high-voltage transmission line. On the other hand, with the rapid development of optical remote sensing technology, the resolution of images acquired by satellites is higher and higher, the detailed information is more and more abundant, and the wide-area monitoring of the running state of the power transmission line of the power grid by utilizing the optical remote sensing image becomes the development direction of remote sensing application research. The method is accurate, but consumes a great deal of manpower and material resources, and is easily influenced by subjective state, experience and the like of people; the other is machine detection, namely, the target characteristics in the image are extracted through a computer, and then the target in the image is detected in a supervision/non-supervision mode by using methods such as statistical pattern recognition, a neural network, a Boosting algorithm, a support vector machine (Support Vector Machines, SVM), an artificial immune system (Artificial Immune System) and the like.
At present, detection and identification of towers by utilizing optical images at home and abroad are mostly based on low-altitude visible light imaging (such as low-altitude unmanned aerial vehicle imaging, vehicle-mounted camera imaging and the like). Under the condition of low-altitude imaging, the target background is mostly sky, the background is cleaner, and the current method mostly requires a specific imaging visual angle to facilitate manual feature extraction, so that the method is difficult to be directly applied to remote sensing image tower detection.
In practice, the conditions of various background types, various imaging angles, weather changes and the like in an imaging scene of a space-based platform or an air-based platform are complex, particularly, when a tower is damaged naturally or artificially, the state of the tower can be changed, the process of invariance characteristics is designed manually, the deep learning-based method can realize that manual intervention is not needed in the characteristic extraction stage, and when a new mode appears, only a corresponding sample retraining model is needed to be added, so that the model has stronger expandability. However, in order to ensure that the detection classification model obtained by deep learning training has higher accuracy, the following assumptions need to be satisfied: 1) The test sample and the training sample are to satisfy independent co-distribution conditions, i.e., i.i.d. (independent andidentically distributed) hypothesis; 2) The number of training samples is sufficiently large. In an actual machine learning application scenario, these two assumptions are difficult to satisfy.
Transfer learning refers to a learning process that applies knowledge learned in an old domain (or called a source domain) to a new domain (or called a target domain) using similarities between data, tasks, or models, and the requirements of the transfer learning process on the sample do not need to satisfy the two assumptions mentioned above. The migration learning method reduces the overfitting risk of the model by sharing knowledge between the source domain and the target domain, so that the detection and identification model can still be used under the condition of insufficient samples.
Disclosure of Invention
Aiming at the problems that in the detection and identification of the high-voltage transmission towers of the remote sensing images, the traditional method has complex manual characteristic design process and is difficult to realize high-precision detection of the positions of the towers in the remote sensing images, and a deep migration learning method is introduced to reduce the requirements of a deep learning algorithm on the target data of the towers, the invention provides an optical remote sensing image tower position detection method based on the deep migration learning. The method is suitable for detecting the position of the transmission tower in the visible light remote sensing image with high accuracy under complex application scenes such as various tower states, background types, illumination and the like, and provides support for safety monitoring of the transmission line.
In order to achieve the above purpose, the technical scheme adopted by the invention is as follows:
an optical remote sensing image tower position detection method based on deep migration learning comprises the following steps:
step one: acquiring optical remote sensing image tower original data and source domain original data, and marking and standardizing to obtain tower data and source domain data;
step two: dividing source domain data into a source domain training set, a source domain verification set and a source domain test set, and dividing tower original data into a target domain training set, a target domain verification set and a target domain test set;
step three: performing data enhancement on the source domain training set and the source domain verification set to obtain source domain training data for network training, and performing data enhancement on the target domain training set and the target domain verification set to obtain target domain training data for network training;
step four: by superposing the acceptance module at the position 2 and adding the RPN module at the rear end of the network, a deep neural network for detecting the tower target is constructed;
step five: training the deep neural network obtained in the fourth step by using the source domain training data obtained in the third step to obtain a detection network on the source domain;
step six: the learning rate is adjusted to be 1/30 of the original learning rate, the classification class number of the network end is 2, and the detection network obtained in the fifth step is subjected to fine adjustment by utilizing the target domain training data obtained in the third step to obtain a detection model on the target domain;
step seven: and D, testing the detection model on the target domain obtained in the step six by using the target domain test set obtained in the step two.
In the third step, data enhancement is performed on the source domain training set, the source domain verification set, the target domain training set and the target domain verification set according to any combination of the following enhancement methods:
1) In order to improve the robustness of the model to the imaging angle, the data of the rotation angle and the turnover are enhanced;
2) In order to improve the robustness of the model to illumination variation, data enhancement for adjusting brightness and saturation is carried out;
3) In order to improve the robustness of the model to the state change of the power transmission tower, data enhancement of distortion is carried out;
4) In order to improve the robustness of the model to the size change of the transmission tower, data enhancement of scaling is carried out.
Further, in the first step, the source domain raw data is a DOTA dataset.
Further, the specific steps of the fourth step are as follows:
and adding an acceptance module at the position 2 in the front half characteristic extraction stage of the neural network, adding an RPN module at the rear end of the network, and connecting the RPN module with the network through a convolution layer or a pooling layer to construct the deep neural network for detecting the pole tower target.
Compared with the prior art, the invention has the beneficial effects that:
(1) Aiming at the problems that the existing method adopting low-altitude visible light imaging (such as low-altitude unmanned aerial vehicle imaging, vehicle-mounted camera imaging and the like) is difficult to apply to optical remote sensing image high-accuracy tower position detection, the traditional optical characteristic manual construction process is complicated, the generalization capability is not strong and the like, the invention provides an optical remote sensing image tower position detection method based on deep migration learning. The method is suitable for detecting the position of the transmission tower in the visible light remote sensing image with high accuracy under complex application scenes such as various tower states, background types, illumination and the like, and provides support for safety monitoring of the transmission line.
(2) Aiming at the problems that in the detection of the position of the optical remote sensing image tower, under complex application scenes such as various tower states, background types, illumination and the like, the image features meeting the requirements on scale and angle, especially the condition that the condition of the tower is kept unchanged and selective, the background of the transmission tower is usually complex and the like, based on the deep learning theory, the invention utilizes the functions of different network layers (convolution layer, pooling layer, regression layer and the like) and combines the functions of special network modules, thereby realizing the automation and abstract expression of the target features under the complex condition and automatically completing the target detection. Compared with the traditional target detection process, the method can greatly improve the accuracy of detecting the tower target in complex application scenes such as various tower states, background types, illumination and the like.
(3) The invention aims to reduce the requirement of the tower target detection depth neural network on the tower target data, realize the training of the tower target detection depth neural network and introduce a deep migration learning method. The existing standard database is used as source domain data, a parameter migration mode is adopted, a detection model trained in the source domain is applied to a target (pole tower) domain, and model parameters are modified in small batches in a fine tuning mode, so that migration of pole tower detection knowledge is realized, and the method has important significance for practical application of a deep learning-based transmission pole tower position detection method.
Drawings
FIG. 1 is a flow chart of a method for detecting the position of a tower of an optical remote sensing image based on deep transfer learning;
FIG. 2 is an example of a VOC standard data format;
FIG. 3 is a block diagram of an acceptance feature extraction module;
FIG. 4 is a block diagram of an RPN recommendation area generation module;
FIG. 5 is a block diagram of a deep neural network for pole tower target detection;
FIG. 6 is a sample graph in the DOTA database;
FIG. 7 is a sample graph in a tower database;
fig. 8 shows the actual image test results.
Detailed Description
The following description of the embodiments and the accompanying drawings 1-8 further illustrate the technical scheme of the present invention, but are not limited thereto, and modifications and equivalents of the technical scheme of the present invention should be included in the protection scope of the present invention without departing from the spirit and scope of the technical scheme of the present invention.
Aiming at the problem that the characteristic process is difficult to design manually in the conventional method in the position detection and identification of the high-voltage transmission tower of the remote sensing image, a detection and identification model is constructed by adopting a deep learning method, and aiming at the problem that a large number of high-quality labeling samples are difficult to obtain in the actual machine learning process, the invention considers the similarity among data, tasks or models in the transfer learning, applies the knowledge learned in the old field (or called a source field) to the new field (or called a target field), introduces the transfer learning method, reduces the requirement of the tower target detection and identification on the tower sample data, and realizes the end-to-end high-accuracy detection and identification of the tower target.
Detailed description of the preferred embodiments
Referring to fig. 1, the embodiment describes an optical remote sensing image tower position detection method based on deep migration learning, and the method comprises the following steps:
step one: acquiring DOTA data as source domain original data, taking transmission line tower image data of a Pleiades satellite as target domain original data, and marking, splitting and data enhancement processing the source domain original data and the target domain original data;
step two: designing a deep neural network structure for detecting towers, and providing a high-accuracy tower position detection neural network model aiming at different tower states, background types, illumination and other scenes;
step three: a deep migration learning method is introduced, the requirement of a model on the image data of the transmission line tower is reduced, and model training is completed;
step four: and testing and verifying the model by using remote sensing images containing four tower states.
Detailed description of the preferred embodiments
The method for detecting the position of the optical remote sensing image tower based on the deep transfer learning in the first embodiment comprises the following specific steps:
DOTA data is selected as source domain original data, transmission line tower image data of a visible light wave band of a Pleiades satellite is selected as target domain original data, and the two parts of data are cut, marked and the like to conform to VOC data formats, and the processed specific formats are shown in figure 2, so that source domain data and target domain data are obtained. Then dividing the source domain data into a source domain training set, a source domain verification set and a source domain test set according to the proportion of 8:1:1, dividing the target domain data into a target domain training set, a target domain verification set and a target domain test set according to the proportion of 8:1:1, and respectively carrying out data enhancement on the source domain training set, the source domain verification set, the target domain training set and the target domain verification set according to any combination of the following enhancement methods:
1) In order to improve the robustness of the model to the imaging angle, the data of the rotation angle and the overturn are enhanced, and the rotation in the range of 0-90 degrees and the vertical or horizontal overturn are randomly carried out;
2) In order to improve the robustness of the model to illumination variation, data enhancement of brightness and saturation adjustment is carried out, and the brightness and saturation are randomly adjusted to be 0.5-1.5 times of the original brightness and saturation;
3) In order to improve the robustness of the model to the state change of the power transmission tower, data enhancement of distortion is carried out, the proportion factors are adopted to randomly take 30-60, and the elastic coefficients randomly take elastic deformation of 4-7 to carry out the distortion;
4) In order to improve the robustness of the model to the size change of the power transmission tower, the scaled data is enhanced, and the random scaled image is 0.2-5 times of the original image.
And carrying out operation of random combination of the enhancement method on each image in the training set, wherein the data size of the training sample after enhancement is about 5000.
Detailed description of the preferred embodiments
The first specific embodiment relates to a method for detecting the position of an optical remote sensing image tower based on deep transfer learning, and the second specific step comprises the following steps:
the neural network acceptance module is mainly characterized in that convolution kernels with different sizes are adopted in the transverse direction of the network, the convolution kernels with different sizes mean that features with different resolutions are extracted, and fusion of features with different scales is carried out by utilizing feature graphs extracted by different receptive field convolution layers. In the actual calculation process, the convolution layer of the large receptive field convolution kernel in the acceptance structure is split into a plurality of small convolution kernel convolution layers, so that the effects of reducing the number of parameters, increasing the nonlinear structure and increasing the network expression capacity are achieved.
The acceptance module is shown in fig. 3, wherein a) is a structural schematic diagram, the module realizes multi-scale feature extraction by using different scale convolution kernels for the same input feature map, b) is a structure adopted in actual network calculation, splits a large scale convolution kernel, introduces a residual structure to reduce the degradation risk of a model, and simultaneously adjusts the depth of the feature map through a 1×1 convolution layer so as to facilitate different scale feature fusion.
The area recommendation network for the present invention makes the recommendation area determination after the network feature extraction structure (the first half of the network), which is also called RPN structure. The RPN structure diagram is shown in fig. 4, and the recommended region generation section removes the region beyond the boundary of the original image, and maps the recommended regions with different sizes into a feature map with a fixed size. The specific implementation process is that according to the size of the output required, the feature map corresponding to the recommended region is partitioned according to the output size, and then each block is subjected to maximum pooling, and finally the output with the fixed size is obtained.
The structure of the deep neural network for optical remote sensing image tower target detection is shown in fig. 5.
For the loss function of the network, if the network model contains only empirical risks, the model is prone to over-fitting. The complexity of the model decision function is a requirement of over-fitting, which needs to be destroyed in order to reduce the risk of over-fitting, i.e. the complexity of the decision function is reduced, and regularization terms can be added to the loss function. Meanwhile, according to the sparsity assumption of the features, namely that only a small number of features are relevant to the task, the method uses L1 regularization as a structural risk item.
The final objective function is as follows:
detailed description of the preferred embodiments
The method for detecting the high precision of the high-resolution optical remote sensing image tower based on the deep migration learning in the first embodiment comprises the following specific steps:
the source domain raw data of the invention selects DOTA (A Larget-scale Dataset for Object Detection inAerial Images) data set, which is provided by the combination of the Wuda remote sensing national heavy laboratory Xia Guisong and the Huake telecom college, and totally comprises 2800 remote sensing images (about 4000X 4000) and about 180000 examples, and the data set is divided into 15 categories. The DOTA data object categories mainly include: baseball fields, swimming pools, large vehicles, airplanes, ships, etc. DOTA datasets are often used as large scale development and testing data for remote sensing image detection algorithms, some examples of which are shown in fig. 6. Compared with the original data of the target domain, the target domain and the target domain are remote sensing data, the bottom characteristic distribution is similar, the ground object categories in the background are similar, such as grasslands, bare soil, water surfaces and the like in the background, and the bottom layer feature categories are similar.
The depth migration training detailed flow of the whole network is as follows:
1) Independently training a network comprising a feature extraction part and an RPN part by using source domain data;
2) Generating a recommended region and corresponding labels by using the network in the step 1) as training samples, extracting partial network parameters by using fixed features, and predicting partial networks by using training positions and categories. At this time, a detection and identification model of the network on the source domain is obtained;
3) And (3) on the basis of the source domain detection model obtained in the step (2), adjusting network training super parameters (reducing the learning rate to be 1/30 of the original) and classifying category numbers (the source domain data has a plurality of types of targets, and the target domain only has one), and performing fine adjustment by utilizing the tower data to finally obtain the detection model on the target domain. The specific fine tuning process is as follows: and replacing the source domain data with the target domain data, and repeating the steps 1) and 2), wherein the model parameters of the step 1) are initialized to the detection model obtained in the step 2) in the source domain, and only the RPN network structure and the subsequent branch network are updated.
Detailed description of the preferred embodiments
The method for detecting the position of the optical remote sensing image tower based on the deep transfer learning in the first embodiment comprises the following specific steps:
and (3) performing target detection by using remote sensing image data containing four tower states and using the detection model obtained in the step (III), wherein fig. 8 is an actual image test result.
Claims (5)
1. The method for detecting the position of the pole tower of the optical remote sensing image based on the deep migration learning is characterized by comprising the following steps of: the method comprises the following steps:
step one: acquiring optical remote sensing image tower original data and source domain original data, and marking and standardizing to obtain tower data and source domain data;
step two: dividing source domain data into a source domain training set, a source domain verification set and a source domain test set, and dividing tower data into a target domain training set, a target domain verification set and a target domain test set;
step three: performing data enhancement on the source domain training set and the source domain verification set to obtain source domain training data for network training, and performing data enhancement on the target domain training set and the target domain verification set to obtain target domain training data for network training;
step four: by superposing the acceptance module at the position 2 and adding the RPN module at the rear end of the network, a deep neural network for detecting the tower target is constructed;
step five: training the deep neural network obtained in the fourth step by using the source domain training data obtained in the third step to obtain a detection network on the source domain;
step six: adjusting network training super parameters and the number of classification categories of the network terminal, and performing fine adjustment on the detection network obtained in the fifth step by utilizing the target domain training data obtained in the third step to obtain a detection model on the target domain;
step seven: and D, testing the detection model on the target domain obtained in the step six by using the target domain test set obtained in the step two.
2. The optical remote sensing image tower position detection method based on deep migration learning of claim 1, wherein the method is characterized by comprising the following steps of: in the third step, data enhancement is performed on the source domain training set, the source domain verification set, the target domain training set and the target domain verification set according to any combination of the following enhancement methods:
1) In order to improve the robustness of the model to the imaging angle, the data of the rotation angle and the turnover are enhanced;
2) In order to improve the robustness of the model to illumination variation, data enhancement for adjusting brightness and saturation is carried out;
3) In order to improve the robustness of the model to the state change of the power transmission tower, data enhancement of distortion is carried out;
4) In order to improve the robustness of the model to the size change of the transmission tower, data enhancement of scaling is carried out.
3. The optical remote sensing image tower position detection method based on deep migration learning of claim 1, wherein the method is characterized by comprising the following steps of: in the first step, the source domain original data is a DOTA data set.
4. The optical remote sensing image tower position detection method based on deep migration learning of claim 1, wherein the method is characterized by comprising the following steps of: the specific steps of the fourth step are as follows:
and adding an acceptance module at the position 2 in the front half characteristic extraction stage of the neural network, adding an RPN module at the rear end of the network, and connecting the RPN module with the network through a convolution layer or a pooling layer to construct the deep neural network for detecting the pole tower target.
5. The optical remote sensing image tower position detection method based on deep migration learning of claim 1, wherein the method is characterized by comprising the following steps of: and step six, fine tuning: the super-parameter learning rate of the network training is adjusted to be 1/30 of the original learning rate, and the classification class number of the network terminal is 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010247180.9A CN111461006B (en) | 2020-03-31 | 2020-03-31 | Optical remote sensing image tower position detection method based on deep migration learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010247180.9A CN111461006B (en) | 2020-03-31 | 2020-03-31 | Optical remote sensing image tower position detection method based on deep migration learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111461006A CN111461006A (en) | 2020-07-28 |
| CN111461006B true CN111461006B (en) | 2023-07-25 |
Family
ID=71682424
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010247180.9A Active CN111461006B (en) | 2020-03-31 | 2020-03-31 | Optical remote sensing image tower position detection method based on deep migration learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111461006B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111898523A (en) * | 2020-07-29 | 2020-11-06 | 电子科技大学 | Remote sensing image special vehicle target detection method based on transfer learning |
| CN112883840B (en) * | 2021-02-02 | 2023-07-07 | 中国人民公安大学 | Power transmission line extraction method based on key point detection |
| CN117132903B (en) * | 2023-10-26 | 2024-01-23 | 江苏云幕智造科技有限公司 | Rotary satellite component identification method based on deep migration learning |
Family Cites Families (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103218810B (en) * | 2013-03-27 | 2016-04-20 | 华北电力大学 | A kind of power tower bar image meaning of one's words dividing method |
| CN204461366U (en) * | 2015-04-03 | 2015-07-08 | 中信戴卡股份有限公司 | A kind of axle cap groove depth on-line measuring device |
| US10192111B2 (en) * | 2017-03-10 | 2019-01-29 | At&T Intellectual Property I, L.P. | Structure from motion for drone videos |
| CN106971200A (en) * | 2017-03-13 | 2017-07-21 | 天津大学 | A kind of iconic memory degree Forecasting Methodology learnt based on adaptive-migration |
| CN107451616A (en) * | 2017-08-01 | 2017-12-08 | 西安电子科技大学 | Multi-spectral remote sensing image terrain classification method based on the semi-supervised transfer learning of depth |
| CN107657279B (en) * | 2017-09-26 | 2020-10-09 | 中国科学院大学 | Remote sensing target detection method based on small amount of samples |
| CN108956640B (en) * | 2018-04-04 | 2020-12-18 | 国网智能科技股份有限公司 | Vehicle-mounted detection device and detection method suitable for power distribution line inspection |
| US10915792B2 (en) * | 2018-09-06 | 2021-02-09 | Nec Corporation | Domain adaptation for instance detection and segmentation |
| CN109332928B (en) * | 2018-10-23 | 2021-07-23 | 江苏山扬智能装备有限公司 | Automatic street lamp post welding system and method based on deep learning online detection |
| CN109977918B (en) * | 2019-04-09 | 2023-05-02 | 华南理工大学 | Target detection positioning optimization method based on unsupervised domain adaptation |
| CN110210355B (en) * | 2019-05-24 | 2021-11-26 | 华南农业大学 | Paddy field weed species identification method and system and target position detection method and system |
| CN110378252A (en) * | 2019-06-28 | 2019-10-25 | 浙江大学 | A kind of distress in concrete recognition methods based on depth migration study |
| CN110780146B (en) * | 2019-12-10 | 2021-04-27 | 武汉大学 | Transformer fault identification and positioning diagnosis method based on multi-stage transfer learning |
-
2020
- 2020-03-31 CN CN202010247180.9A patent/CN111461006B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN111461006A (en) | 2020-07-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109584248B (en) | Infrared target instance segmentation method based on feature fusion and dense connection network | |
| WO2022111219A1 (en) | Domain adaptation device operation and maintenance system and method | |
| CN110033002B (en) | License plate detection method based on multitask cascade convolution neural network | |
| CN111461006B (en) | Optical remote sensing image tower position detection method based on deep migration learning | |
| Chen et al. | Vehicle detection in high-resolution aerial images based on fast sparse representation classification and multiorder feature | |
| CN110796009A (en) | Method and system for detecting marine vessel based on multi-scale convolution neural network model | |
| CN104517103A (en) | Traffic sign classification method based on deep neural network | |
| CN110969171A (en) | Image classification model, method and application based on improved convolutional neural network | |
| CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
| CN110852327A (en) | Image processing method, image processing device, electronic equipment and storage medium | |
| CN113591617B (en) | Deep learning-based water surface small target detection and classification method | |
| CN111882620A (en) | Road drivable area segmentation method based on multi-scale information | |
| CN115272306B (en) | Solar cell panel grid line enhancement method utilizing gradient operation | |
| CN113269224A (en) | Scene image classification method, system and storage medium | |
| CN116385958A (en) | Edge intelligent detection method for power grid inspection and monitoring | |
| CN116091946A (en) | Yolov 5-based unmanned aerial vehicle aerial image target detection method | |
| Li et al. | An aerial image segmentation approach based on enhanced multi-scale convolutional neural network | |
| CN116152678A (en) | Marine disaster-bearing body identification method based on twin neural network under small sample condition | |
| CN111695460A (en) | Pedestrian re-identification method based on local graph convolution network | |
| Gu et al. | A classification method for polsar images using SLIC superpixel segmentation and deep convolution neural network | |
| CN110796716B (en) | Image coloring method based on multiple residual error network and regularized transfer learning | |
| CN116883650A (en) | Image-level weak supervision semantic segmentation method based on attention and local stitching | |
| Liu et al. | Integration transformer for ground-based cloud image segmentation | |
| Huang et al. | Research on crop planting area classification from remote sensing image based on deep learning | |
| CN115661451A (en) | Deep learning single-frame infrared small target high-resolution segmentation method |
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 |