CN113283366A - Building side information extraction method based on learning vector quantization algorithm - Google Patents
Building side information extraction method based on learning vector quantization algorithm Download PDFInfo
- Publication number
- CN113283366A CN113283366A CN202110630283.8A CN202110630283A CN113283366A CN 113283366 A CN113283366 A CN 113283366A CN 202110630283 A CN202110630283 A CN 202110630283A CN 113283366 A CN113283366 A CN 113283366A
- Authority
- CN
- China
- Prior art keywords
- side information
- learning
- neuron
- vector quantization
- layer
- 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
- 239000013598 vector Substances 0.000 title claims abstract description 68
- 238000000605 extraction Methods 0.000 title claims abstract description 27
- 238000013139 quantization Methods 0.000 title claims abstract description 27
- 210000002569 neuron Anatomy 0.000 claims abstract description 49
- 238000005070 sampling Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 15
- 238000013528 artificial neural network Methods 0.000 claims abstract description 14
- 238000000034 method Methods 0.000 claims description 18
- 238000011156 evaluation Methods 0.000 claims description 6
- 239000011159 matrix material Substances 0.000 claims description 6
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000002372 labelling Methods 0.000 claims description 4
- 238000003909 pattern recognition Methods 0.000 abstract description 3
- 230000008569 process Effects 0.000 description 4
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000003595 spectral 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses a building side information extraction method based on a learning vector quantization algorithm, which comprises the steps of constructing a learning vector quantization neural network training set; setting sampling parameters and characteristic vectors; initializing a weight and a learning rate; calculating the Euclidean distance; recording neuron class labels of an output layer; updating the weight value of the class label; outputting a side information clustering result; the invention adopts a learning vector quantization algorithm to realize the building side information extraction method, and uses a small amount of weight vectors to represent the topological structure of data, so that the LVQ algorithm has wider application in the field of pattern recognition, and meanwhile, the extraction precision can be rapidly improved in the building side information extraction.
Description
Technical Field
The invention relates to the technical field of building information extraction, in particular to a method for extracting building side information based on a learning vector quantization algorithm.
Background
With the development of the multisource remote sensing technology, high-resolution remote sensing images in the same region and different time phases can be quickly and conveniently acquired at the present stage, the images have higher-resolution ground feature information and richer ground feature spectral information, space data support is provided for extracting building side information by utilizing the information, a learning vector quantization algorithm is a supervised neural network classification method with simple structure and strong function, the supervised neural network classification method is successfully applied to multiple fields of statistics, pattern recognition, machine learning and the like as a nearest neighbor prototype classifier, LVQ is used as a nearest neighbor prototype classifier through continuously updating a neuron weight vector (prototype) in the training process and continuously adjusting the learning rate of the neuron weight vector (prototype), the boundary between different types of weight vectors can be gradually converged to a Bayesian classification boundary, and in the algorithm, the selection of a winning neuron (nearest neighbor weight vector) is judged by calculating the Euclidean distance between an input sample and the weight vector, the most outstanding advantage of the LVQ is that the LVQ has self-adaptability, and a codebook of training samples can be obtained in an online learning mode;
at present, in the process of obtaining the side information of the building, the side information of the building is extracted through a high-resolution remote sensing image, so that the accuracy of information extraction in the extraction process is low, and meanwhile, the side information of the building is extracted through a clustering algorithm and manual extraction of characteristic information, so that the operation is complex, the accuracy of information extraction cannot be ensured, and the existing extraction mode cannot meet the requirements of popularization and application at present.
Disclosure of Invention
The invention provides a method for extracting building side information based on a learning vector quantization algorithm, which can effectively solve the problems that the accuracy of information extraction in the extraction process is low because the building side information is extracted by a high-resolution remote sensing image singly in the process of obtaining the building side information, the building side information is extracted by a clustering algorithm and manually extracting characteristic information, the operation is complicated, the accuracy of information extraction cannot be ensured, and the existing extraction mode cannot meet the requirements of popularization and application at present.
In order to achieve the purpose, the invention provides the following technical scheme: a building side information extraction method based on a learning vector quantization algorithm specifically comprises the following extraction steps:
s1, constructing a learning vector quantization neural network training set;
s2, setting sampling parameters and feature vectors;
s3, initializing a weight and a learning rate;
s4, calculating the Euclidean distance;
s5, recording neuron labels of an output layer;
s6, updating the weight value of the class label;
s7, outputting a side information clustering result;
and S8, sampling and evaluating accuracy.
According to the technical scheme, in the S1, the step of constructing the learning vector quantization neural network training set refers to the step of preprocessing the building high-resolution remote sensing image data of the information to be extracted, after preprocessing, the step of performing target image cutting on the remote sensing image data information of the building, and establishing building category labels;
during class labeling, the side face of the building is set as a foreground face, other faces of the building are set as background faces, and meanwhile, a single-channel gray image is output to serve as a network training set.
According to the above technical solution, in S2, the setting of the sampling parameters and the feature vectors means that the original picture and the tagged picture are circularly input according to the pixel interval and the window size of the determined block, the number of blocks is calculated at the same time, the positive and negative sample quantities are calculated in units of blocks, and then the positive and negative sample quantities are balanced by uniformly sampling the samples.
According to the above technical solution, in S3, the initialization weight and the learning rate specifically refer to a weight and a learning rate between an initialization input layer and a competition layer, the input layer is directly connected to the output layer, each node of the output layer has a weight vector connected to it, and the purpose of learning is to find an optimal value of the weight.
According to the above technical solution, in S4, the calculating the euclidean distance specifically includes adding the input vector to the input layer to calculate the euclidean distance between the neuron in the competition layer and the input vector;
the euclidean distance is a commonly used distance definition, and refers to the true distance between two points in a multidimensional space, or the natural length of a vector, i.e., the distance from the point to the origin, and the euclidean distance in two-dimensional and three-dimensional spaces is the actual distance between the two points.
According to the above technical solution, in S5, the recording of the output layer neuron class label specifically refers to selecting a competition layer neuron with a minimum distance from an input vector, where each neuron of the input layer corresponds to an input feature, and a plurality of neurons arranged in parallel correspond to input multidimensional feature vectors, and the input layer is fully interconnected with the hidden layer by a weight, that is, each neuron of the input layer is connected to each neuron of the hidden layer, the multidimensional feature vectors and the plurality of hidden layer neurons determine a weight matrix, and each row vector corresponds to one neuron of the hidden layer;
and a plurality of neurons of the hidden layer form distribution in a multidimensional characteristic space to form a classified class center, and each neuron of the hidden layer corresponds to an output layer neuron representing the final classification through a weight matrix and is recorded as an output layer neuron class label connected with the neuron.
According to the above technical solution, in S6, updating the weight of the class label means updating the weight of the class label corresponding to the input vector, and the neural network continuously adjusts the connection weight by learning the training sample data, thereby continuously adjusting the learning rate.
According to the above technical solution, in S7, outputting the side information clustering result specifically means outputting the side information clustering result after the specified learning rate and iteration number and the straight road reaches the preset iteration number requirement or precision requirement.
According to the technical scheme, in S8, the sampling evaluation accuracy specifically refers to an accuracy of the LVQ learning evaluated in a sampling manner, an evaluation threshold is determined first when the accuracy is evaluated, and after the accuracy meets a threshold requirement, a complete building side information result is output in the area, so that the building side information extraction of the whole area is completed.
Compared with the prior art, the invention has the beneficial effects that:
the invention adopts a learning vector quantization algorithm to realize the building side information extraction method, utilizes a small amount of weight vectors to represent the topological structure of data, compared with an unsupervised self-organizing neural network algorithm, because a supervision signal is introduced in the weight vector updating process, the LVQ algorithm has wider application in the field of pattern recognition, and simultaneously can quickly improve the extraction precision in the building side information extraction.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the principles of the invention and not to limit the invention.
In the drawings:
FIG. 1 is a flow chart of the steps of the extraction method of the present invention;
fig. 2 is a diagram of a learning vector quantization neural network structure of the present invention.
Detailed Description
The preferred embodiments of the present invention will be described in conjunction with the accompanying drawings, and it will be understood that they are described herein for the purpose of illustration and explanation and not limitation.
Example (b): as shown in fig. 1-2, the present invention provides a technical solution, a method for extracting building side information based on a learning vector quantization algorithm, which specifically includes the following extraction steps:
s1, constructing a learning vector quantization neural network training set;
s2, setting sampling parameters and feature vectors;
s3, initializing a weight and a learning rate;
s4, calculating the Euclidean distance;
s5, recording neuron labels of an output layer;
s6, updating the weight value of the class label;
s7, outputting a side information clustering result;
and S8, sampling and evaluating accuracy.
Based on the technical scheme, in S1, the step of constructing the learning vector quantization neural network training set is to preprocess the building high-resolution remote sensing image data of the information to be extracted, perform target image cutting on the remote sensing image data information of the building after preprocessing, and establish building category labeling;
during class labeling, the side face of the building is set as a foreground face, other faces of the building are set as background faces, and meanwhile, a single-channel gray image is output to serve as a network training set.
Based on the above technical solution, in S2, setting the sampling parameters and the feature vectors means that the original picture and the tagged picture are circularly input according to the pixel interval and the window size of the determined block, the number of blocks is calculated at the same time, the positive and negative sample quantities are calculated by using the blocks as units, and then the positive and negative sample quantities are balanced by uniformly sampling the samples.
Based on the above technical solution, in S3, initializing the weight and the learning rate specifically refer to initializing the weight and the learning rate between the input layer and the competition layer, where the input layer is directly connected to the output layer, each node in the output layer has a weight vector connected to it, and the purpose of learning is to find the optimal value of the weight.
Based on the above technical solution, in S4, calculating the euclidean distance specifically means adding the input vector to the input layer to calculate the euclidean distance between the neuron in the competition layer and the input vector;
the euclidean distance is a commonly used distance definition, and refers to the true distance between two points in a multidimensional space, or the natural length of a vector, i.e., the distance from the point to the origin, and the euclidean distance in two-dimensional and three-dimensional spaces is the actual distance between the two points.
Based on the above technical solution, in S5, recording the output layer neuron class label specifically means selecting a competition layer neuron with the minimum distance from the input vector, where each neuron of the input layer corresponds to an input feature, and a plurality of neurons arranged in parallel correspond to input multidimensional feature vectors, and the input layer is fully interconnected with the hidden layer by a weight, that is, each neuron of the input layer is connected to each neuron of the hidden layer, the multidimensional feature vectors and the plurality of hidden layer neurons determine a weight matrix, and each row vector corresponds to one neuron of the hidden layer;
and a plurality of neurons of the hidden layer form distribution in a multidimensional characteristic space to form a classified class center, and each neuron of the hidden layer corresponds to an output layer neuron representing the final classification through a weight matrix and is recorded as an output layer neuron class label connected with the neuron.
Based on the above technical solution, in S6, updating the weight of the class label means updating the weight according to the class label corresponding to the input vector, and the neural network continuously adjusts the connection weight by learning the training sample data, thereby continuously adjusting the learning rate.
Based on the above technical solution, in S7, outputting the side information clustering result specifically means outputting the side information clustering result after the specified learning rate and iteration number and the straight road reaches the preset iteration number requirement or precision requirement.
Based on the above technical scheme, in S8, the sampling evaluation accuracy specifically refers to an accuracy of the LVQ learning evaluated in a sampling manner, an evaluation threshold is determined first when the accuracy is evaluated, and after the accuracy meets a threshold requirement, a complete building side information result is output in this area, so as to achieve the purpose of extracting the building side information of the whole area.
Finally, it should be noted that: although the present invention has been described in detail with reference to the foregoing embodiments, it will be apparent to those skilled in the art that changes may be made in the embodiments and/or equivalents thereof without departing from the spirit and scope of the invention. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (9)
1. A building side information extraction method based on a learning vector quantization algorithm is characterized by comprising the following steps: the method specifically comprises the following extraction steps:
s1, constructing a learning vector quantization neural network training set;
s2, setting sampling parameters and feature vectors;
s3, initializing a weight and a learning rate;
s4, calculating the Euclidean distance;
s5, recording neuron labels of an output layer;
s6, updating the weight value of the class label;
s7, outputting a side information clustering result;
and S8, sampling and evaluating accuracy.
2. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in the step S1, the step of constructing the learning vector quantization neural network training set refers to the step of preprocessing the building high-resolution remote sensing image data of the information to be extracted, and after preprocessing, the step of performing target image cutting on the remote sensing image data information of the building and establishing building category labels;
during class labeling, the side face of the building is set as a foreground face, other faces of the building are set as background faces, and meanwhile, a single-channel gray image is output to serve as a network training set.
3. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in S2, setting the sampling parameters and the eigenvectors refers to circularly inputting the original picture and the tagged picture according to the determined pixel interval and window size of the block, calculating the number of blocks, calculating the positive and negative sample quantities in units of blocks, and then balancing the positive and negative sample quantities by uniformly sampling the samples.
4. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in S3, the initialization weight and the learning rate specifically refer to a weight and a learning rate between an initialization input layer and a competition layer, the input layer is directly connected to the output layer, each node of the output layer has a weight vector connected to it, and the purpose of learning is to find an optimal value of the weight.
5. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in S4, calculating the euclidean distance specifically includes adding the input vector to the input layer to calculate the euclidean distance between the neuron in the competition layer and the input vector;
the euclidean distance is a commonly used distance definition, and refers to the true distance between two points in a multidimensional space, or the natural length of a vector, i.e., the distance from the point to the origin, and the euclidean distance in two-dimensional and three-dimensional spaces is the actual distance between the two points.
6. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in S5, recording the output layer neuron class label specifically means selecting a competition layer neuron with the smallest distance from the input vector, where each neuron of the input layer corresponds to an input feature, and a plurality of neurons arranged in parallel correspond to input multidimensional feature vectors, and the input layer is fully interconnected with the hidden layer by a weight, that is, each neuron of the input layer is connected to each neuron of the hidden layer, and the multidimensional feature vectors and the plurality of hidden layer neurons determine a weight matrix, and each row vector corresponds to one neuron of the hidden layer;
and a plurality of neurons of the hidden layer form distribution in a multidimensional characteristic space to form a classified class center, and each neuron of the hidden layer corresponds to an output layer neuron representing the final classification through a weight matrix and is recorded as an output layer neuron class label connected with the neuron.
7. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in S6, updating the weight value of the class label means updating the weight value according to the class label corresponding to the input vector, and the neural network continuously adjusts the connection weight value by learning the training sample data, thereby continuously adjusting the learning rate.
8. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in S7, outputting the side information clustering result specifically means outputting the side information clustering result after the specified learning rate and iteration count and the straight track reaches a preset iteration count requirement or precision requirement.
9. The method for extracting the building side information based on the learning vector quantization algorithm according to claim 1, wherein: in S8, the sampling evaluation accuracy specifically refers to an accuracy of the LVQ learning evaluated in a sampling manner, an evaluation threshold is determined first when the accuracy is evaluated, and after the accuracy meets a threshold requirement, a complete building side information result is output in this area, so as to complete the extraction of the building side information in the whole area.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110630283.8A CN113283366A (en) | 2021-06-07 | 2021-06-07 | Building side information extraction method based on learning vector quantization algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110630283.8A CN113283366A (en) | 2021-06-07 | 2021-06-07 | Building side information extraction method based on learning vector quantization algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113283366A true CN113283366A (en) | 2021-08-20 |
Family
ID=77283592
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110630283.8A Pending CN113283366A (en) | 2021-06-07 | 2021-06-07 | Building side information extraction method based on learning vector quantization algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113283366A (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110069972A (en) * | 2017-12-11 | 2019-07-30 | 赫克斯冈技术中心 | Automatic detection real world objects |
-
2021
- 2021-06-07 CN CN202110630283.8A patent/CN113283366A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110069972A (en) * | 2017-12-11 | 2019-07-30 | 赫克斯冈技术中心 | Automatic detection real world objects |
Non-Patent Citations (3)
| Title |
|---|
| 宋旭东等: "基于高分遥感影像的建筑物侧面信息提取及其高度计算", 《北京测绘》 * |
| 谢丽蓉等: "基于LVQ-GA-BP神经网络的煤矿瓦斯涌出量预测", 《煤矿安全》 * |
| 黄冰晶等: "基于LVQ神经网络的建筑物提取", 《软件导刊》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111047182B (en) | Airspace complexity evaluation method based on deep unsupervised learning | |
| CN108388927B (en) | Small sample polarization SAR terrain classification method based on deep convolution twin network | |
| CN108537264B (en) | Heterogeneous image matching method based on deep learning | |
| CN111814871A (en) | Image classification method based on reliable weight optimal transmission | |
| CN112101430B (en) | Anchor frame generation method for image target detection processing and lightweight target detection method | |
| CN112347970B (en) | Remote sensing image ground object identification method based on graph convolution neural network | |
| CN109741341B (en) | Image segmentation method based on super-pixel and long-and-short-term memory network | |
| CN107992850B (en) | Outdoor scene three-dimensional color point cloud classification method | |
| CN110619059B (en) | Building marking method based on transfer learning | |
| CN110414616B (en) | Remote sensing image dictionary learning and classifying method utilizing spatial relationship | |
| CN106874862B (en) | Crowd counting method based on sub-model technology and semi-supervised learning | |
| CN108549866B (en) | Remote sensing airplane identification method based on dense convolutional neural network | |
| CN112052772A (en) | Face shielding detection algorithm | |
| CN112800876A (en) | Method and system for embedding hypersphere features for re-identification | |
| CN111311702B (en) | Image generation and identification module and method based on BlockGAN | |
| CN113313164A (en) | Digital pathological image classification method and system based on superpixel segmentation and image convolution | |
| CN111259917B (en) | Image feature extraction method based on local neighbor component analysis | |
| CN114841257A (en) | Small sample target detection method based on self-supervision contrast constraint | |
| CN112364791A (en) | Pedestrian re-identification method and system based on generation of confrontation network | |
| CN114842264A (en) | Hyperspectral image classification method based on multi-scale spatial spectral feature joint learning | |
| CN112488128A (en) | Bezier curve-based detection method for any distorted image line segment | |
| CN114692732A (en) | Method, system, device and storage medium for updating online label | |
| CN113408505B (en) | Chromosome polarity identification method and system based on deep learning | |
| CN104598898A (en) | Aerially photographed image quick recognizing system and aerially photographed image quick recognizing method based on multi-task topology learning | |
| CN114283326A (en) | Underwater target re-identification method combining local perception and high-order feature reconstruction |
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: 20210820 |
|
| RJ01 | Rejection of invention patent application after publication |