CN110569884A - Hyperspectral remote sensing image classification method based on deep learning and morphology - Google Patents
Hyperspectral remote sensing image classification method based on deep learning and morphology Download PDFInfo
- Publication number
- CN110569884A CN110569884A CN201910756938.9A CN201910756938A CN110569884A CN 110569884 A CN110569884 A CN 110569884A CN 201910756938 A CN201910756938 A CN 201910756938A CN 110569884 A CN110569884 A CN 110569884A
- Authority
- CN
- China
- Prior art keywords
- hyperspectral
- remote sensing
- spectral
- deep learning
- image obtained
- 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.)
- Withdrawn
Links
Classifications
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
Abstract
The invention discloses a hyperspectral remote sensing image classification method based on deep learning and morphology, which comprises the following steps of: s1: adopting a smoothing filter with three gradually increasing structural element sizes to carry out multi-scale smoothing filtering on the hyperspectral images of each wave band on the spatial dimension and blend spatial information into spectral information; s2: performing top hat transformation on the hyperspectral image obtained in the step S1, and correcting the influence of uneven illumination; s3: performing zero-averaging on the hyperspectral image obtained in the step S2; s4: performing dimension conversion on the hyperspectral image obtained in the step S3; s5: extracting spectral features; s6: the spectral features obtained in step S5 are classified. The invention can improve the reliability and simplify the classification process.
Description
Technical Field
The invention relates to a hyperspectral remote sensing image processing technology, in particular to a hyperspectral remote sensing image classification method based on deep learning and morphology.
Background
After the 60's of the 20 th century, the united states achieved widespread use of multispectral technology, and began research into the spectral characteristics of terrestrial objects. In the 80 s of the 20 th century, the appearance of hyperspectral technology has brought the field of remote sensing into the hyperspectral remote sensing era. Through the development of the last 40 years, the hyperspectral technology has very wide application in the aspect of earth observation. The hyperspectral remote sensing image has abundant spatial information and spectral information, the image data acquired by the imaging spectrometer is a 'data cube', and the fine classification of ground objects becomes possible due to massive data. The hyperspectral remote sensing image classification is widely and effectively applied to the fields of agriculture, environment, geology and the like, and great economic benefits and ecological benefits are generated. In recent years, the development trend of deep learning is met, the remote sensing field is more widely developed on the basis of the technology, and the accuracy of hyperspectral image classification is greatly improved.
Traditional hyperspectral image classification methods, such as Support Vector Machine (SVM), K-nearest neighbor (KNN), K-means clustering (K-means) and other methods, can only learn shallow features of mass data and are seriously affected by dimension disaster, so that higher classification accuracy cannot be obtained. Although the deep learning is long in time consumption compared with the traditional classification method, more abstract characteristics of data can be learned, and the classification task can be completed more accurately. Aiming at the classification problem of hyperspectral images, a plurality of researchers at home and abroad deeply explore and provide a plurality of optimization and classification algorithms based on the traditional method and deep learning. Among them, the convolutional neural network is prominent in image classification. In 2012, professor Hinton and students constructed deep convolutional neural networks, and a method for reducing weight attenuation amplitude is adopted to prevent training model fitting. This improved algorithm greatly increased the computing power of computers and achieved good performance over the Imagenet race, and this step forward in image recognition research. At present, the convolutional neural network has been successfully applied to the aspects of face recognition, face detection, natural image recognition, image classification and the like.
Through the exploration of countless researchers, the current hyperspectral remote sensing image classification strategies are mainly divided into two main categories: the classification based on spectral vector information only, and the combined classification of spectral-spatial information. However, classification based on spectral vector information only and spectral-spatial information joint classification tend to be complex in process and poor in reliability.
disclosure of Invention
The purpose of the invention is as follows: the invention aims to provide a hyperspectral remote sensing image classification method based on deep learning and morphology, which can solve the technical problems of complex classification process and poor reliability in the prior art.
The technical scheme is as follows: the invention relates to a hyperspectral remote sensing image classification method based on deep learning and morphology, which comprises the following steps of:
S1: adopting a smoothing filter with three gradually increasing structural element sizes to carry out multi-scale smoothing filtering on the hyperspectral images of each wave band on the spatial dimension and blend spatial information into spectral information;
S2: performing top hat transformation on the hyperspectral image obtained in the step S1, and correcting the influence of uneven illumination;
S3: performing zero-averaging on the hyperspectral image obtained in the step S2;
S4: performing dimension conversion on the hyperspectral image obtained in the step S3;
S5: extracting spectral features;
s6: the spectral features obtained in step S5 are classified.
Further, the step S2 specifically includes the following steps: and (4) opening the hyperspectral image obtained in the step (S1) by using the structural element, and subtracting the hyperspectral image after the opening operation from the hyperspectral image obtained in the step (S1), thereby correcting the influence of uneven illumination.
Further, the step S3 specifically includes the following steps: and (4) performing zero-averaging on the spectral vector of each pixel point in the hyperspectral image obtained in the step (S2), namely subtracting an average gray value from the gray value of each waveband of each pixel point.
Further, the step S4 specifically includes the following steps: and converting the one-dimensional spectral vectors of the pixel points in the hyperspectral image obtained in the step S3 into two-dimensional spectral vectors.
Further, in step S5, the spectral features are extracted using four convolutional layers, in which 5 × 5 and 3 × 3 convolution kernels are used.
Further, in step S6, the spectral features are classified by using a classifier composed of two fully-connected layers and a Softmax layer.
has the advantages that: the invention discloses a hyper-spectral remote sensing image classification method based on deep learning and morphology, which is characterized in that a smoothing filter with three gradually-increased structural element sizes is used for carrying out multi-scale smoothing filtering on hyper-spectral images of each wave band in a spatial dimension, spatial information is integrated into spectral information, noise can be removed in the spatial dimension, the spatial correlation of ground feature distribution is enhanced, the spectral information is corrected in the spectral dimension, the reliability of the spectral information for classification is improved, and extra extraction of the spatial information is not needed like a spectrum-spatial information combined classification method in the prior art, so that the classification process can be simplified while the reliability is improved.
Drawings
FIG. 1 is a flow chart of a method in accordance with an embodiment of the present invention;
FIG. 2 is a schematic diagram of a morphological filter according to an embodiment of the present invention;
FIG. 3 is a diagram of a convolutional neural network model for extracting spectral features in an embodiment of the present invention.
Detailed Description
the specific embodiment discloses a hyperspectral remote sensing image classification method based on deep learning and morphology, as shown in fig. 1, the hyperspectral remote sensing image classification method comprises the following steps:
S1: adopting a smoothing filter with three gradually increasing structural element sizes to carry out multi-scale smoothing filtering on the hyperspectral images of each wave band on the spatial dimension and blend spatial information into spectral information;
s2: performing top hat transformation on the hyperspectral image obtained in the step S1, and correcting the influence of uneven illumination; fig. 2 is a schematic diagram of a morphological filter comprising three smoothing filters in step S1, and further comprising means capable of performing a top-hat transform in step S2.
s3: performing zero-averaging on the hyperspectral image obtained in the step S2;
S4: performing dimension conversion on the hyperspectral image obtained in the step S3;
s5: extracting spectral features;
s6: the spectral features obtained in step S5 are classified.
Step S2 specifically includes the following processes: and (4) opening the hyperspectral image obtained in the step (S1) by using the structural element, and subtracting the hyperspectral image after the opening operation from the hyperspectral image obtained in the step (S1), thereby correcting the influence of uneven illumination.
Step S3 specifically includes the following processes: and (4) performing zero-averaging on the spectral vector of each pixel point in the hyperspectral image obtained in the step (S2), namely subtracting an average gray value from the gray value of each waveband of each pixel point.
Step S4 specifically includes the following processes: and converting the one-dimensional spectral vectors of the pixel points in the hyperspectral image obtained in the step S3 into two-dimensional spectral vectors.
In step S5, a convolutional neural network model as shown in fig. 3 is used to extract spectral features, where the convolutional neural network model has four convolutional layers, and the convolutional layers use convolution kernels of 5 × 5 and 3 × 3. And performing batch normalization by adopting a batch normalization algorithm before extracting the spectral features. Through the processing of the activation function RELU, the output of a part of neurons is 0, so that the network is sparse, and the overfitting problem is relieved.
In step S6, the spectral features are classified using a classifier composed of two fully-connected layers and a Softmax layer.
In order to verify the method, three kinds of hyperspectral data of Indian pipes, University of Pavia and Salinas are adopted for classification and testing.
Indian Pines imaging was performed in 1992 by the american airborne visible infrared spectrometer (AVIRIS), and was the earliest remote sensing image used for hyperspectral classification testing. The spatial dimension of the Indian Pines hyperspectral image is 145 multiplied by 145, the spatial resolution is 20m, the number of wave bands is 220, and the spectral range is 0.4-2.5 μm. The image is an image of an agricultural land in indiana, usa, and since there is a lot of noise in the bands of 104 to 108 and 150 to 163, 200 bands excluding the 20 bands are generally used to classify the ground surface coverings into 16 types, and the image is used for agricultural research.
TABLE 1 Indian Pines high spectral data ground object type situation
The University of Pavia hyperspectral image was imaged in 2003, the imaging area was Pavea city in Italy, and the adopted device was a German airborne reflective optical spectral imager. The spectrum imager has 115 wave bands, the spectrum range is 0.43-0.86 mu m, and the space size is 610 multiplied by 340. Wherein 12 wave bands are removed by the noise image, the remaining 103 wave bands can be used for research, and the wave bands are totally divided into 9 types of ground objects.
TABLE 2 University of Pavia hyperspectral data ground object class situation
The Salinas hyperspectral image was also imaged using an American airborne visible Infrared Spectroscopy (AVIRIS), the imaging region being located in California, USA. Except that the spatial resolution of the image is 3.7 m, the size is 512 x 127, the number of bands is 204, and the image is divided into 16 types of ground objects.
TABLE 3 Salinas Hyperspectral data ground object class situation
In order to evaluate the performance of the classification method proposed by the present invention, tables 4 and 5 are the experimental results of the training data sets Indian Pines and University of Pavia of the method proposed by the present invention (MF + Spe _ CNN), respectively, compared to the methods proposed by some researchers.
TABLE 4 comparison of Indian Pines classifications with other methods
TABLE 5 comparison of University of Pavia classifications with other methods
Denoising and comparing: (1) gaussian filtering has been used to preprocess spectral images; (2) gao Xin et al adopted improved adaptive diffusion coefficients and joint edge detection to denoise images and then use DBN to realize classification; (3) and the like, the classification is realized by using KNN after noise points of images of each wave band are removed by using a recursive filtering algorithm. The three classification methods are all denoising firstly and then finish the classification of Indian pins hyperspectral images by applying spectral information, and the table 4 shows that the OA, AA and Kappa obtained by the method for morphological filtering denoising and spectral feature extraction classification provided by the invention are higher than the three methods, which shows that the method of the invention has better performance than other methods using denoising treatment.
And (3) classification strategy comparison: (1) tan et al propose a classification method for realizing space-spectrum combination by adding spatial features by using an EMP technology; (2) the process and the like provide a classification method combining space spectrum information with active deep learning; (3) the Zhu and the like adopt a null spectrum combined classification method, a ranking method is used for fusing to obtain rotationally invariant null spectrum features, and then a deep belief network is used for completing classification. (1) The EMP technology is used, namely morphology is expanded, and spatial information is added through the morphology; (2) the classification method (3) also uses spatial information in combination with spectral information to realize the classification of the University of Pavia hyperspectral images, and it can be seen from table 5 that the method of the present invention has the highest precision compared with the methods (1), (2) and (3). From the principle of morphological operation, it can be considered that spatial information is merged into spectral information when a filter denoises, so that space-spectrum integration is realized.
Compared with the traditional method: shen et al, Lu et al are based on the optimized classification of the traditional method, and have a large difference compared with most of the hyperspectral remote sensing image classification methods based on deep learning, wherein the Kappa coefficient of the method provided by the invention is higher than 10 percentage points compared with the SVM method adopted by Shen et al, Lu et al. In conclusion, in the face of huge data volume of the hyperspectral image, the deep learning can fully utilize massive data to learn abstract features, and the classification is easier.
through classification and testing of three hyperspectral data of Indian pipes, University of Pavia and Salinas, experimental results show that the hyperspectral classification method provided by the invention has certain effectiveness, removes noise, corrects spectral information, greatly improves average classification precision and overall classification precision, achieves Kappa coefficients of more than 0.9, and has highly similar edge characteristics of ground feature distribution.
Claims (6)
1. the hyperspectral remote sensing image classification method based on deep learning and morphology is characterized by comprising the following steps of: the method comprises the following steps:
S1: adopting a smoothing filter with three gradually increasing structural element sizes to carry out multi-scale smoothing filtering on the hyperspectral images of each wave band on the spatial dimension and blend spatial information into spectral information;
s2: performing top hat transformation on the hyperspectral image obtained in the step S1, and correcting the influence of uneven illumination;
S3: performing zero-averaging on the hyperspectral image obtained in the step S2;
S4: performing dimension conversion on the hyperspectral image obtained in the step S3;
S5: extracting spectral features;
S6: the spectral features obtained in step S5 are classified.
2. The hyperspectral remote sensing image classification method based on deep learning and morphology according to claim 1 is characterized in that: the step S2 specifically includes the following steps: and (4) opening the hyperspectral image obtained in the step (S1) by using the structural element, and subtracting the hyperspectral image after the opening operation from the hyperspectral image obtained in the step (S1), thereby correcting the influence of uneven illumination.
3. the hyperspectral remote sensing image classification method based on deep learning and morphology according to claim 1 is characterized in that: the step S3 specifically includes the following steps: and (4) performing zero-averaging on the spectral vector of each pixel point in the hyperspectral image obtained in the step (S2), namely subtracting an average gray value from the gray value of each waveband of each pixel point.
4. The hyperspectral remote sensing image classification method based on deep learning and morphology according to claim 1 is characterized in that: the step S4 specifically includes the following steps: and converting the one-dimensional spectral vectors of the pixel points in the hyperspectral image obtained in the step S3 into two-dimensional spectral vectors.
5. the hyperspectral remote sensing image classification method based on deep learning and morphology according to claim 1 is characterized in that: in step S5, the spectral features are extracted using four convolutional layers, in which 5 × 5 and 3 × 3 convolutional kernels are used.
6. The hyperspectral remote sensing image classification method based on deep learning and morphology according to claim 1 is characterized in that: in step S6, a classifier composed of two fully-connected layers and a Softmax layer is used to classify the spectral features.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910756938.9A CN110569884A (en) | 2019-08-16 | 2019-08-16 | Hyperspectral remote sensing image classification method based on deep learning and morphology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910756938.9A CN110569884A (en) | 2019-08-16 | 2019-08-16 | Hyperspectral remote sensing image classification method based on deep learning and morphology |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110569884A true CN110569884A (en) | 2019-12-13 |
Family
ID=68775622
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910756938.9A Withdrawn CN110569884A (en) | 2019-08-16 | 2019-08-16 | Hyperspectral remote sensing image classification method based on deep learning and morphology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110569884A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111881953A (en) * | 2020-07-14 | 2020-11-03 | 安徽大学 | Remote sensing hyperspectral image classification method based on local binary pattern and KNN classifier |
| CN111898423A (en) * | 2020-06-19 | 2020-11-06 | 北京理工大学 | Morphology-based multisource remote sensing image ground object fine classification method |
| CN112580480A (en) * | 2020-12-14 | 2021-03-30 | 河海大学 | Hyperspectral remote sensing image classification method and device |
| CN112733775A (en) * | 2021-01-18 | 2021-04-30 | 苏州大学 | Hyperspectral image classification method based on deep learning |
| CN112818831A (en) * | 2021-01-29 | 2021-05-18 | 河南大学 | Hyperspectral image classification algorithm based on band clustering and improved domain transformation recursive filtering |
-
2019
- 2019-08-16 CN CN201910756938.9A patent/CN110569884A/en not_active Withdrawn
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111898423A (en) * | 2020-06-19 | 2020-11-06 | 北京理工大学 | Morphology-based multisource remote sensing image ground object fine classification method |
| CN111898423B (en) * | 2020-06-19 | 2022-12-23 | 北京理工大学 | Morphology-based multisource remote sensing image ground object fine classification method |
| CN111881953A (en) * | 2020-07-14 | 2020-11-03 | 安徽大学 | Remote sensing hyperspectral image classification method based on local binary pattern and KNN classifier |
| CN111881953B (en) * | 2020-07-14 | 2022-04-22 | 安徽大学 | Remote sensing hyperspectral image classification method based on local binary pattern and KNN classifier |
| CN112580480A (en) * | 2020-12-14 | 2021-03-30 | 河海大学 | Hyperspectral remote sensing image classification method and device |
| CN112580480B (en) * | 2020-12-14 | 2024-03-26 | 河海大学 | Hyperspectral remote sensing image classification method and device |
| CN112733775A (en) * | 2021-01-18 | 2021-04-30 | 苏州大学 | Hyperspectral image classification method based on deep learning |
| CN112818831A (en) * | 2021-01-29 | 2021-05-18 | 河南大学 | Hyperspectral image classification algorithm based on band clustering and improved domain transformation recursive filtering |
| CN112818831B (en) * | 2021-01-29 | 2022-09-16 | 河南大学 | Hyperspectral image classification algorithm based on band clustering and improved domain transformation recursive filtering |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110569884A (en) | Hyperspectral remote sensing image classification method based on deep learning and morphology | |
| Liu et al. | A 3-D atrous convolution neural network for hyperspectral image denoising | |
| CN112233026A (en) | SAR image denoising method based on multi-scale residual attention network | |
| CN111160273B (en) | Hyperspectral image spatial spectrum joint classification method and device | |
| He et al. | A dual global–local attention network for hyperspectral band selection | |
| CN112580480B (en) | Hyperspectral remote sensing image classification method and device | |
| CN111667019B (en) | Hyperspectral image classification method based on deformable separation convolution | |
| Ye et al. | Hyperspectral image classification using principal components-based smooth ordering and multiple 1-D interpolation | |
| He et al. | Kernel low-rank multitask learning in variational mode decomposition domain for multi-/hyperspectral classification | |
| CN104881682A (en) | Image classification method based on locality preserving mapping and principal component analysis | |
| Zhang et al. | Hyperspectral image classification using spatial and edge features based on deep learning | |
| CN107292855A (en) | A kind of image de-noising method of the non local sample of combining adaptive and low-rank | |
| Zhou et al. | RGB-to-HSV: A Frequency-Spectrum Unfolding Network for Spectral Super-Resolution of RGB Videos | |
| CN113052130A (en) | Hyperspectral image classification method based on depth residual error network and edge protection filtering | |
| Li et al. | Ga-cnn: Convolutional neural network based on geometric algebra for hyperspectral image classification | |
| Song et al. | Using dual-channel CNN to classify hyperspectral image based on spatial-spectral information | |
| Sabrol et al. | Recognition of tomato late blight by using DWT and component analysis | |
| Lv et al. | Edge protection filtering and convolutional neural network for hyperspectral remote sensing image classification | |
| Shi et al. | A complementary spectral-spatial method for hyperspectral image classification | |
| CN117115675A (en) | Cross-time-phase light-weight spatial spectrum feature fusion hyperspectral change detection method, system, equipment and medium | |
| Hamouda et al. | Modified convolutional neural network based on adaptive patch extraction for hyperspectral image classification | |
| Menaka et al. | Classification of multispectral satellite images using sparse SVM classifier | |
| CN111860654B (en) | Hyperspectral image classification method based on cyclic neural network | |
| CN112132104B (en) | ISAR ship target image domain enhancement identification method based on loop generation countermeasure network | |
| Liu et al. | Target detection of hyperspectral image based on convolutional neural networks |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20191213 |
|
| WW01 | Invention patent application withdrawn after publication |