CN113537033A - Building rubbish remote sensing image identification method based on deep learning - Google Patents
Building rubbish remote sensing image identification method based on deep learning Download PDFInfo
- Publication number
- CN113537033A CN113537033A CN202110785190.2A CN202110785190A CN113537033A CN 113537033 A CN113537033 A CN 113537033A CN 202110785190 A CN202110785190 A CN 202110785190A CN 113537033 A CN113537033 A CN 113537033A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sensing image
- network
- data set
- model
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A building rubbish remote sensing image recognition method based on deep learning belongs to the field of remote sensing image recognition. The existing remote sensing image identification method is easy to interfere, and the whole information can not be mined, so that the identification precision is low. A building rubbish remote sensing image recognition method based on deep learning comprises the steps of preprocessing an obtained remote sensing image to obtain a remote sensing image data set; expanding a remote sensing image data set sample, adding an L2 regularization punishment item to the seventh layer of the neural network, and training a network model added with an L2 regularization punishment item by using the expanded data set to obtain a target identification model; the mIOU ratio of the intersection and the union of the real value and the predicted value in the semantic segmentation method of deep Lab is calculated to realize the improvement of the semantic segmentation algorithm; and image recognition is carried out by utilizing an improved recognition model and an algorithm. The method identifies the accuracy draft, can monitor the processing progress of illegal stacking, and realizes dynamic tracking monitoring and purification of the urban environment.
Description
Technical Field
The invention relates to a building rubbish remote sensing image identification method based on deep learning.
Background
The remote sensing image identification roughly goes through the following processes: traditional remote sensing image identification methods based on pixel, such as maximum likelihood method and K-Means mean value method, but the image spectrum brightness information is easy to be interfered, and the whole information can not be mined, so that the method is easy to generate 'salt and pepper noise', and is only used as a contrast item or a preprocessing method at present; based on the object-oriented remote sensing identification method, although the advantage of rich attribute features of the polygonal object is exerted, the object is easy to over-segment or under-segment, and the segmentation scale is not easy to determine; the image semantics based on the image elements are segmented into popular research directions for the current remote sensing image recognition, the characteristics of strong self-learning capability and fault-tolerant capability of the image semantics based on the image elements are derived from deep learning, and the research and implementation of thousands of classification methods are established.
However, the effectiveness of the deep learning method also depends on the richness of the training data, and a large amount of sample data becomes a necessary condition for research. The sample expansion can be obtained by using a simple data enhancement method such as inversion, and can also be obtained by using a machine learning method such as generation countermeasure network to perform picture synthesis. In addition, the deep learning related research method often requires that the feature distribution of training data and test data is similar, but the requirement is difficult to achieve in practical project application, so the deep learning method is difficult to be applied in engineering projects due to the data requirement. Migration learning is a branch of deep learning research methods and is a great research hotspot at present.
Disclosure of Invention
The invention aims to solve the problems that the existing remote sensing image identification method is easy to interfere and low in identification precision due to the fact that the existing remote sensing image identification method cannot mine whole information, and provides a building rubbish remote sensing image identification method based on deep learning.
A building rubbish remote sensing image identification method based on deep learning is achieved through the following steps:
firstly, preprocessing an acquired remote sensing image to obtain a remote sensing image data set;
expanding a remote sensing image data set sample, adding an L2 regularization punishment item into the seventh layer of the neural network, and training a network model added with an L2 regularization punishment item by using the expanded data set to obtain a target identification model;
step three, improving a semantic segmentation algorithm by calculating the mIOU ratio of the intersection and the union of the real value and the predicted value in the semantic segmentation method of deep Lab;
and step four, carrying out image recognition by using the improved recognition model and the improved algorithm.
Preferably, the step of preprocessing the acquired remote sensing image to obtain a remote sensing image data set in the step one includes:
and performing remote sensing image preprocessing operation of orthorectification and image fusion on the remote sensing image by adopting an ENVI platform, and performing histogram equalization operation on the result data.
Preferably, the step of expanding the remote sensing image dataset sample described in the step two fuses features of a plurality of images in a manner of improving generation of a countermeasure network, and the specific steps include:
the GAN generates new data on the basis of the original data set, and the GAN generation countermeasure network comprises two models: generating a model and a discrimination model, wherein the representative symbols of the two models are G and D respectively; the game implementation using these two models generates a competing network,
and if z is random noise and x is real data, the generative network and the discriminant network are respectively represented by G and D, wherein D can be regarded as a two-classifier, and then the cross entropy is adopted for representation, and the description is as follows:
minmaxV＝Ex～pdata(x)[logD(x)]+Ez～pz(z)[log(1-D(G(z)))]
wherein, logD (x) of the first item represents the judgment of the discriminator on the real data, and log (1-D (G (z)) of the second item represents the synthesis and judgment of the data; through such a Max-min game, G and D are respectively optimized to train the required generating network and the required discriminating network circularly and alternately until a Nash equilibrium point is reached;
the DCGAN provides a training framework on the basis of the GAN, the DCGAN conducts training guidance on the GAN, replaces a full connection layer with a convolution layer, removes a pooling layer, and introduces development results of a discriminant model into a generated model by adopting Batch Normalization (BN) technology.
Preferably, the step three of improving the recognition model and the recognition semantic segmentation algorithm by calculating the mliou ratio of the intersection and the union of the real value and the predicted value in the semantic segmentation method of deep lab includes:
a semantic segmentation method of deep Lab is adopted, the condition randomness and the convolution neural network are combined, the full convolution network is used as a basis, and each layer is continuously optimized; then, the mIOU ratio of the intersection and union of the real value and the predicted value is calculated, and the improvement of the semantic segmentation algorithm is realized.
The invention has the beneficial effects that:
the invention applies the deep learning algorithm to the semantic segmentation of the remote sensing image and to the identification of the construction waste in the remote sensing image, thereby saving more manpower and material resources. The method is characterized in that a sample expansion experiment for image generation is carried out aiming at the problem of few samples of a building remote sensing data set, urban building rubbish is detected from the aspect of semantic segmentation of remote sensing images, a reliable data source expansion method is provided for urban building rubbish remote sensing monitoring, and technical support is provided for building rubbish stockpiling management. Meanwhile, the earth observation remote sensing technology has the characteristics of long-distance detection, large-area coverage, short revisit period and the like, whether the construction waste is cleared or not can be found quickly through the research, the current situation information such as the stacking area of the construction waste can be mastered, the illegal stacking processing progress can be monitored, and the urban environment can be dynamically tracked, monitored and purified.
Drawings
FIG. 1 is a flow chart of a method of the present invention;
FIG. 2 is a basic model for generating a countermeasure network to which the present invention relates;
FIG. 3 is a diagram of a DCGAN generator according to the present invention;
FIG. 4 shows the structure of DeepLabv3ASPP according to the present invention.
Detailed Description
The first embodiment is as follows:
in the embodiment, as shown in fig. 1, the method for identifying the remote sensing image of the construction waste based on deep learning is implemented by the following steps:
firstly, preprocessing an acquired remote sensing image to obtain a remote sensing image data set;
expanding a remote sensing image data set sample, adding an L2 regularization punishment item into the seventh layer of the neural network, and training a network model added with an L2 regularization punishment item by using the expanded data set to obtain a target identification model;
step three, improving a semantic segmentation algorithm by calculating the mIOU ratio of the intersection and the union of the real value and the predicted value in the semantic segmentation method of deep Lab;
and step four, carrying out image recognition by using the improved recognition model and the improved algorithm.
The second embodiment is as follows:
different from the first specific embodiment, in the first method for identifying remote sensing images of construction waste based on deep learning of the present embodiment, the step of preprocessing the acquired remote sensing images to obtain a remote sensing image data set includes:
due to the influence of the overall positioning precision of a remote sensing information platform and the error rate of a sensor, the reason that layers of panchromatic images and multispectral images in a satellite remote sensing technology image can not be aligned and the like, more preprocessing operations are needed to be carried out on the satellite remote sensing image, the remote sensing image preprocessing operations of orthorectification and image fusion are carried out on the remote sensing image by adopting an ENVI platform, and histogram equalization operation is carried out on the result data; the relative position precision of the remote sensing data is improved, the image quality is improved, and the data characteristics are enhanced.
The third concrete implementation mode:
different from the first or second specific embodiments, in the method for identifying the remote sensing images of the construction waste based on deep learning of the second embodiment, the step of expanding the remote sensing image data set sample fuses the features of the plurality of images in a mode of improving and generating a countermeasure network, and the specific steps include:
the method is characterized in that a sample expansion experiment of image generation is carried out aiming at the problem of few samples of a building remote sensing data set, urban building rubbish is detected from the aspect of semantic segmentation of remote sensing images, a reliable data source expansion method is provided for urban building rubbish remote sensing monitoring, and technical support is provided for building rubbish stockpiling management. The generation of the confrontation network is improved, the semantic segmentation network precision is improved, and the production image is closer to a real image.
Common data expansion is implemented by flipping, randomly cropping, rotating, locally distorting the image, and using GAN (creating a competing network) methods. Training of artificial intelligence requires a large number of data sets, which can be costly if collected and labeled all by human labor.
The GAN generates new data on the basis of the original data set so as to train a more robust model, and the GAN generation countermeasure network comprises two models: generating a model and a discrimination model, wherein the representative symbols of the two models are G and D respectively; the game using the two models realizes the generation of the countermeasure network, thereby leading the two models to improve the overall competition effect.
And if z is random noise and x is real data, the generative network and the discriminant network are respectively represented by G and D, wherein D can be regarded as a two-classifier, and then the cross entropy is adopted for representation, and the description is as follows:
minmaxV＝Ex～pdata(x)[logD(x)]+Ez～pz(z)[log(1-D(G(z)))]
wherein, logD (x) of the first item represents the judgment of the discriminator on the real data, and log (1-D (G (z)) of the second item represents the synthesis and judgment of the data; through such a Max-min game, G and D are respectively optimized to train the required generating network and the required discriminating network circularly and alternately until a Nash equilibrium point is reached; generating a basic model of the countermeasure network is shown in FIG. 2;
the DCGAN provides a training framework on the basis of the GAN, the DCGAN conducts training guidance on the GAN, replaces a full connection layer with a convolution layer, removes a pooling layer, and introduces development results of a discriminant model into a generated model by adopting Batch Normalization (BN) technology. The structure of the DCGAN generator is shown in fig. 3. In addition, DCGAN also emphasizes the importance and guidance of hidden layer analysis and visual counting on GAN training
The fourth concrete implementation mode:
different from the third specific embodiment, in the method for identifying a building waste remote sensing image based on deep learning of the third embodiment, the step of improving the identification model and the identification semantic segmentation algorithm by calculating the mliou ratio of the intersection and the union of the real value and the predicted value in the semantic segmentation method of deep lab includes:
by adopting a semantic segmentation method of deep Lab, by combining conditional randomness and a convolutional neural network, and using a full convolutional network as a basis, each layer is continuously optimized, so that the change of a receptive field can be ensured in the semantic segmentation process, and the position information in the aspect of space can also be reserved; then, the mIOU ratio of the intersection and union of the real value and the predicted value is calculated, and the improvement of the semantic segmentation algorithm is realized, so that the accuracy is increased.
The Deeplab can extract different features and control different resolutions, an algorithm has a unique solution, and each level is continuously optimized by combining conditional randomness and a convolutional neural network and using a full convolutional network as a basis, so that the change of a receptive field can be ensured in a semantic segmentation process, and the position information in the aspect of space can be also reserved. The DeepLabv3ASPP structure is shown in FIG. 4.
The above is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made to the present invention by those skilled in the art. 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 (4)
1. A building rubbish remote sensing image identification method based on deep learning is characterized in that: the method is realized by the following steps:
firstly, preprocessing an acquired remote sensing image to obtain a remote sensing image data set;
expanding a remote sensing image data set sample, adding an L2 regularization punishment item into the seventh layer of the neural network, and training a network model added with an L2 regularization punishment item by using the expanded data set to obtain a target identification model;
step three, improving a semantic segmentation algorithm by calculating the mIOU ratio of the intersection and the union of the real value and the predicted value in the semantic segmentation method of deep Lab;
and step four, carrying out image recognition by using the improved recognition model and the improved algorithm.
2. The building rubbish remote sensing image recognition method based on deep learning of claim 1, characterized in that: the step one of preprocessing the acquired remote sensing image to obtain a remote sensing image data set comprises the following steps:
and performing remote sensing image preprocessing operation of orthorectification and image fusion on the remote sensing image by adopting an ENVI platform, and performing histogram equalization operation on the result data.
3. The construction waste remote sensing image recognition method based on deep learning according to claim 1 or 2, characterized in that: step two the step of expanding the remote sensing image data set sample fuses the characteristics of a plurality of images in a mode of improving and generating a countermeasure network, and the specific steps comprise:
the GAN generates new data on the basis of the original data set, and the GAN generation countermeasure network comprises two models: generating a model and a discrimination model, wherein the representative symbols of the two models are G and D respectively; the game implementation using these two models generates a competing network,
and if z is random noise and x is real data, the generative network and the discriminant network are respectively represented by G and D, wherein D can be regarded as a two-classifier, and then the cross entropy is adopted for representation, and the description is as follows:
minmaxV＝Ex～pdata(x)[logD(x)]+Ez～pz(z)[log(1-D(G(z)))]
wherein, logD (x) of the first item represents the judgment of the discriminator on the real data, and log (1-D (G (z)) of the second item represents the synthesis and judgment of the data; through such a Max-min game, G and D are respectively optimized to train the required generating network and the required discriminating network circularly and alternately until a Nash equilibrium point is reached;
the DCGAN provides a training framework on the basis of the GAN, the DCGAN conducts training guidance on the GAN, replaces a full connection layer with a convolution layer, removes a pooling layer, and introduces development results of a discriminant model into a generated model by adopting Batch Normalization (BN) technology.
4. The building rubbish remote sensing image recognition method based on deep learning of claim 3 is characterized in that: step three, the step of improving the recognition model and the recognition semantic segmentation algorithm by calculating the mIOU ratio of the intersection and the union of the real value and the predicted value in the semantic segmentation method of deep Lab comprises the following steps:
a semantic segmentation method of deep Lab is adopted, the condition randomness and the convolution neural network are combined, the full convolution network is used as a basis, and each layer is continuously optimized; then, the mIOU ratio of the intersection and union of the real value and the predicted value is calculated, and the improvement of the semantic segmentation algorithm is realized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110785190.2A CN113537033A (en) | 2021-07-12 | 2021-07-12 | Building rubbish remote sensing image identification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110785190.2A CN113537033A (en) | 2021-07-12 | 2021-07-12 | Building rubbish remote sensing image identification method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113537033A true CN113537033A (en) | 2021-10-22 |
Family
ID=78127465
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110785190.2A Pending CN113537033A (en) | 2021-07-12 | 2021-07-12 | Building rubbish remote sensing image identification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113537033A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115147729A (en) * | 2022-06-27 | 2022-10-04 | 浙江大学 | Solid waste landfill site risk identification method based on sky ground data and intelligent algorithm |
| CN116091953A (en) * | 2023-04-11 | 2023-05-09 | 耕宇牧星(北京)空间科技有限公司 | Building rubbish identification method based on grouping wavelet calibration network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110598784A (en) * | 2019-09-11 | 2019-12-20 | 北京建筑大学 | Machine learning-based construction waste classification method and device |
| CN111027445A (en) * | 2019-12-04 | 2020-04-17 | 安徽工程大学 | Target identification method for marine ship |
| CN112489054A (en) * | 2020-11-27 | 2021-03-12 | 中北大学 | Remote sensing image semantic segmentation method based on deep learning |
| CN112861752A (en) * | 2021-02-23 | 2021-05-28 | 东北农业大学 | Crop disease identification method and system based on DCGAN and RDN |
-
2021
- 2021-07-12 CN CN202110785190.2A patent/CN113537033A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110598784A (en) * | 2019-09-11 | 2019-12-20 | 北京建筑大学 | Machine learning-based construction waste classification method and device |
| CN111027445A (en) * | 2019-12-04 | 2020-04-17 | 安徽工程大学 | Target identification method for marine ship |
| CN112489054A (en) * | 2020-11-27 | 2021-03-12 | 中北大学 | Remote sensing image semantic segmentation method based on deep learning |
| CN112861752A (en) * | 2021-02-23 | 2021-05-28 | 东北农业大学 | Crop disease identification method and system based on DCGAN and RDN |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115147729A (en) * | 2022-06-27 | 2022-10-04 | 浙江大学 | Solid waste landfill site risk identification method based on sky ground data and intelligent algorithm |
| CN116091953A (en) * | 2023-04-11 | 2023-05-09 | 耕宇牧星(北京)空间科技有限公司 | Building rubbish identification method based on grouping wavelet calibration network |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Chen et al. | Vehicle detection in high-resolution aerial images via sparse representation and superpixels | |
| Sánchez-Rodríguez et al. | Automated detection and decomposition of railway tunnels from Mobile Laser Scanning Datasets | |
| CN112232199A (en) | Wearing mask detection method based on deep learning | |
| CN113537033A (en) | Building rubbish remote sensing image identification method based on deep learning | |
| CN113689445B (en) | High-resolution remote sensing building extraction method combining semantic segmentation and edge detection | |
| Kahraman et al. | Road extraction techniques from remote sensing images: A review | |
| CN113505670A (en) | Remote sensing image weak supervision building extraction method based on multi-scale CAM and super-pixels | |
| Babawuro et al. | Satellite imagery cadastral features extractions using image processing algorithms: A viable option for cadastral science | |
| Yang et al. | C-RPNs: Promoting object detection in real world via a cascade structure of Region Proposal Networks | |
| Tran et al. | Pp-linknet: Improving semantic segmentation of high resolution satellite imagery with multi-stage training | |
| CN110929670A (en) | Muck truck cleanliness video identification and analysis method based on yolo3 technology | |
| Xiao et al. | 3D urban object change detection from aerial and terrestrial point clouds: A review | |
| Yang et al. | Superpixel image segmentation-based particle size distribution analysis of fragmented rock | |
| CN105469099B (en) | Pavement crack detection and identification method based on sparse representation classification | |
| Yadav et al. | Building change detection using multi-temporal airborne LiDAR data | |
| Su et al. | Which CAM is better for extracting geographic objects? A perspective from principles and experiments | |
| Sharifi et al. | Efficiency evaluating of automatic lineament extraction by means of remote sensing (Case study: Venarch, Iran) | |
| Wang et al. | Pavement crack detection using attention u-net with multiple sources | |
| Wolf et al. | Applicability of neural networks for image classification on object detection in mobile mapping 3d point clouds | |
| Xu et al. | Crowd density estimation based on improved Harris & OPTICS Algorithm | |
| Chen et al. | All-in-one YOLO architecture for safety hazard detection of environment along high-speed railway | |
| Böhm et al. | Diskmask: Focusing object features for accurate instance segmentation of elongated or overlapping objects | |
| Li et al. | Evaluation of super-resolution on bird detection performance based on deep convolutional networks | |
| Yan et al. | Building Extraction at Amodal-Instance-Segmentation Level: Datasets and Framework | |
| Purohit et al. | ConeQuest: A Benchmark for Cone Segmentation on Mars |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20211022 |
|
| WD01 | Invention patent application deemed withdrawn after publication |