Vegetation Classification of Alpine Grassland in Qinghai Lake Basin Based on HSI hyperspectral remote sensing data
OPTOSKYBLOGBlogsHyperspec Blog
Vegetation Classification of Alpine Grassland in Qinghai Lake Basin Based on HSI hyperspectral remote sensing data
Vegetation Classification of Alpine Grassland in Qinghai Lake Basin Based on HSI hyperspectral remote sensing data
author: Gavin
2022-01-07

The Qinghai Lake Basin is located in the northeastern part of the Qinghai-Tibet Plateau. It is a typical arid area, with extremely fragile ecosystems. The types and distribution of vegetation within the territory have always been a difficult problem for research. The altitude of this area is generally higher (above 3200m), with a larger area and diverse vegetation types. Compared with traditional field surveys and multispectral visual interpretation, hyperspectral remote sensing data has a wide coverage, more bands and The ten rich features of spectral information have significant advantages in the fine classification of vegetation in this area.
Principle
Three dimensionality reduction methods of MNF, KPCA and ISOMAP are used to reduce the dimensionality of the image. By analyzing the respective regularized feature maps after dimensionality reduction, it is concluded that ISOMAP has obvious advantages in information concentration when processing HSI data. The advantages and disadvantages of the two dimensionality reduction methods of ISOMAP based on traditional Euclidean distance and IMED-ISOMAP based on image Euclidean distance. It is believed that there is not much difference between the two methods in terms of information and average gradient, but the ISOMAP algorithm based on traditional Euclidean distance High efficiency, suitable for dimensionality reduction application of HSI in Qinghai Lake basin. For image classification after dimensionality reduction, by comparing the accuracy of three classification algorithms, maximum likelihood, artificial neural network and support vector machine (SVM), it is believed that SVM is superior to the other two in the classification of hyperspectral data. ISOMAP+SVM is more suitable for the production of grassland vegetation classification maps in the Qinghai Lake basin.
Solution
Spectral collection uses ATH9010, ATH9010 is low cost hyperspectral camera for industrial application, It it employs CMOS sensor and the spectral range is 380-1000nm.
The vegetation spectrum range studied in this paper happens to be included in the spectrum range selected by the HSI sensor design, and because of its unique advantages of high spectral resolution, this paper uses HSI hyperspectral data for research. According to the requirements of the national standard remote sensing image plan production specification (1995) for image selection, the HSI data with better vegetation growth season quality in the study area was screened, and the scene image was finally selected as the research data as shown in Figure 1.
Figure 1.
Due to the limitation of the altitude of the study area and the survey time, this survey only sampled relatively low altitude areas in the area surrounding the Qinghai Lake and obtained about 160 samples, as shown in Figure 2. These samples include vegetation such as Kobresia and Elymus vulgaris, mixed vegetation such as Stipa and Bupleurum, a small amount of noxious weeds such as Pedicularis edulis, and vegetation such as Prunus vulgaris, which is relatively distributed at a higher altitude.


Figure.3
Principle
Three dimensionality reduction methods of MNF, KPCA and ISOMAP are used to reduce the dimensionality of the image. By analyzing the respective regularized feature maps after dimensionality reduction, it is concluded that ISOMAP has obvious advantages in information concentration when processing HSI data. The advantages and disadvantages of the two dimensionality reduction methods of ISOMAP based on traditional Euclidean distance and IMED-ISOMAP based on image Euclidean distance. It is believed that there is not much difference between the two methods in terms of information and average gradient, but the ISOMAP algorithm based on traditional Euclidean distance High efficiency, suitable for dimensionality reduction application of HSI in Qinghai Lake basin. For image classification after dimensionality reduction, by comparing the accuracy of three classification algorithms, maximum likelihood, artificial neural network and support vector machine (SVM), it is believed that SVM is superior to the other two in the classification of hyperspectral data. ISOMAP+SVM is more suitable for the production of grassland vegetation classification maps in the Qinghai Lake basin.
Solution
Spectral collection uses ATH9010, ATH9010 is low cost hyperspectral camera for industrial application, It it employs CMOS sensor and the spectral range is 380-1000nm.
- Perform basic preprocessing of image data.
- Use three dimensionality reduction methods of MNF, KPCA and ISOMAP to reduce the dimensionality of the image. By analyzing the respective regularized feature maps after dimensionality reduction, it is concluded that ISOMAP is processing HSI Data has obvious advantages in information concentration.
- On the basis of the superiority of ISOMAP, the applicability of ISOMAP based on traditional Euclidean distance and IMED-ISOMAP based on image Euclidean distance is compared.
- Three classification methods: maximum likelihood, artificial neural network and SVM are used to classify the reduced-dimensional images, and the classification accuracy is calculated to show that SVM is superior to the other two in the classification of hyperspectral data.
- Apply the obtained ISOMAP and SVM combination method to the entire Qinghai Lake basin to obtain the vegetation type distribution map of the Qinghai Lake basin.
The vegetation spectrum range studied in this paper happens to be included in the spectrum range selected by the HSI sensor design, and because of its unique advantages of high spectral resolution, this paper uses HSI hyperspectral data for research. According to the requirements of the national standard remote sensing image plan production specification (1995) for image selection, the HSI data with better vegetation growth season quality in the study area was screened, and the scene image was finally selected as the research data as shown in Figure 1.

Figure 1.

Figure 2
Vegetation maps are divided into vegetation zoning maps and vegetation type maps. The vegetation zoning map shows the regional distribution and distribution law of the combination of vegetation classification units. Its mapping unit, namely the vegetation zoning unit, is complete and continuous on the map; the mapping unit of the vegetation type map, namely the vegetation classification unit, is scattered on the earth. , Is discontinuous on the graph, and what the graph reflects is its actual distribution and regularity of distribution. Vegetation map is a concise and vivid way of expressing the results of vegetation research. The level of vegetation classification unit it can represent, the level of detail and the distribution law, mainly depend on the scale. Generally speaking, the smaller the scale, the higher the level of the vegetation classification unit, the rougher the content, the more general the vegetation distribution law reflected, and vice versa.


Figure.3
Conclusion
Using ISOMAP dimensionality reduction and SVM classification methods to process 31 scenes of HSI hyperspectral remote sensing data in the Qinghai Lake Basin, supplemented with data from field surveys and high spatial resolution images, the production of the vegetation type distribution map of the Qinghai Lake Basin was completed. Compared with the 1:1 million vegetation map, the distribution map of vegetation types in the Qinghai Lake basin has added three new vegetation types that were previously ignored due to small distribution areas or mixed growth: Elymus dahurica, Bupleurum and Pedicularis. The number of land patches was increased from 190 to 13,690, the boundaries of each taxon were adjusted, and the distribution area of each taxon was refined, which better reflected the detailed characteristics of the spatial distribution of vegetation in the Qinghai Lake basin.
Related products
Hyperspectral remote sensing data is characterized by its wide coverage, many bands and rich spectral information.
Hyperspectral Camera:ATH1010
Airborne Hyperspectral Remote Sensing System:ATH9010
Drone Hyperspectral Imaging System:ATHL9010
Related articles
Using ISOMAP dimensionality reduction and SVM classification methods to process 31 scenes of HSI hyperspectral remote sensing data in the Qinghai Lake Basin, supplemented with data from field surveys and high spatial resolution images, the production of the vegetation type distribution map of the Qinghai Lake Basin was completed. Compared with the 1:1 million vegetation map, the distribution map of vegetation types in the Qinghai Lake basin has added three new vegetation types that were previously ignored due to small distribution areas or mixed growth: Elymus dahurica, Bupleurum and Pedicularis. The number of land patches was increased from 190 to 13,690, the boundaries of each taxon were adjusted, and the distribution area of each taxon was refined, which better reflected the detailed characteristics of the spatial distribution of vegetation in the Qinghai Lake basin.
Related products
Hyperspectral remote sensing data is characterized by its wide coverage, many bands and rich spectral information.
Hyperspectral Camera:ATH1010
Airborne Hyperspectral Remote Sensing System:ATH9010
Drone Hyperspectral Imaging System:ATHL9010
Related articles
- Belkin M, Niyogi P. Laplacian eigenmaps fbr dimensionality reduction and data representation]J]. Neural Computation ,2003,Vol. 15(6): 1373-1396
- Boser B E, Guyon I M, Vapnik V N. A training algorithm for optimal margin classifiers[C]. Proceedings of the fifth annual workshop on Computational learning theory ACM, 1992
- Bregler C, Omohundro S M. Nonlinear manifold learning fbr visual speech recognition[C]. Proceedings of the Fifth International Conference on Computer Vision. IEEE, 1995
- Chang C I. Hyperspectral imaging:Techniques fbr spectral detection and classification[M]. New York: Plenum,2003.
Related News
Precision fertilization by UAV for rice at tillering stage in cold region based on hyperspectral remote sensing prescription map
2023-02-01 363New product|recommendation-GF900 UAV-borne laser methane telemetry system
2023-08-30 43Application of UAV-hyperspectral imaging for rice growth monitoring
2023-01-18 502Application of hyperspectral camera in industrial inspection
2023-01-17 331Estimation Scheme of Rice Yield Based on UAV Hyperspectral Images
2023-01-17 278Surveillance of pine wood nematode disease based on satellite remote sensing images
2023-01-16 306UAV forest fire patrol protection plan
2023-01-16 312Application of Hyperspectral Imager in Disguised Target Recognition(part 2)
2023-01-13 348Application of hyperspectral imager in detection of exogenous pests in jujube fruit
2023-01-11 369Application of UAV Hyperspectral in Garbage Sorting
2023-01-10 333Application of Hyperspectral Imager in Disguised Target Recognition(part 1)
2023-01-10 300What is the hyperspectral imaging?
2023-01-05 326Study on the degree of damage to the Asian carlocust with hyperspectral remote sensing model
2023-01-03 294Nitrogen detection in cotton leaves based on hyperspectral
2022-12-23 283FAQ_HYPERSPECTAL IMAGER FAQ
2022-08-22 943Using ATHL9010 to estimate above-ground carbon stocks at the single-tree scale in subtropical forests
2022-02-25 1077Airborne Hyperspectral Imaging in Early Monitoring of Pine Wood Nematode
2022-01-14 868A Study on Eutrophication of Lake Based on Hyper-spectraI Remote
2022-01-07 622Application of Hyperspectral in the Construction of Hongshan Polymetallic Prospecting Model
2022-01-07 576Research on low-contrast wounded target search technology based on hyperspectrum
2022-01-07 555