Hyperspectral Remote Sensing is the acquisition of data from the Earth’s surface and atmosphere using airborne sensors or space borne sensors. The main difference of Hyperspectral Remote Sensing from MultiSpectral Remote Sensing is the ability of HyperSpectral Remote sensing to capture information across a wide range of ElectroMagnetic Spectrum. Unlike traditional Sensor that records only a few broad spectrum bands, hyperspectral sensors gather data across hundreds of narrow bands.This capability allows the creation of the continuous spectral reflectance signatures, making the hyperspectral data useful for in-depth analysis of Earth’s surface.
The data that are collected via the hyperspectral remote sensing is organized into hypercubes and each cube contains two spatial dimensions and one spectral dimension. These hypercubes are the building blocks to extract valuable information from the image.
Hyperspectral data cube structure
The cube consists of four main components:
Push-Broom Scanning
The Push Broom Scanner or Along-Track Scanner is a spectroscopic sensor that is commonly used in Remote sensing from space or airborne. It operates by continuously scanning along a specific path, offering advantage in light capture but has potentially lower resolution compared to other methods.
Sequential Scan Lines
In the above image, multiple scan lines are stacked on top of another creating a 3 dimensional hyperspectral data cube. The X and Y axis in the image represent the spatial data, whereas the Z dimension represents the Spectral information.
Stacked Spatial Images
Each 2 Dimensional Spatial image corresponds to a specific narrow waveband and hyper spectral data cubes typically contain hundreds of stacked images forming a hyperspectral data cube.
Spectral Samples
These samples allow the detection and classification of features in the spectra, serving as the primary mechanism for scene analysis.
In obtaining the HyperSpectral Data, there are 3 different methods that can be employed based on the type of imaging spectrometers. They are
- dispersive elements-based approach,
- spectral filters-based approach,
- snapshot hyperspectral imaging
These methods are used to capture hyperspectral images with varying spatial and temporal resolutions with sensors mounted on different platforms such as unmanned aerial vehicle ( UAV ), airplanes and close-range platforms.
Hyperspectral Imaging Platforms Comparison
| Criteria | Platform A | Platform B | Platform C |
|---|---|---|---|
| Wavelength Range | 400 – 1000 nm | 450 – 900 nm | 380 – 1100 nm |
| Spatial Resolution | 0.5 meters | 1 meter | 0.3 meters |
| Spectral Resolution | 10 nm | 5 nm | 3 nm |
| Number of Bands | 200 | 300 | 400 |
| Imaging Speed | 1000 lines/second | 500 lines/second | 1500 lines/second |
| Data Storage and Transfer | 1 TB SSD, USB 3.0 | 500 GB HDD, USB 2.0 | 2 TB SSD, Thunderbolt 3 |
| Operating Temperature Range | -10°C to 40°C | -20°C to 50°C | -5°C to 45°C |
| Integration with UAV/Drone | Yes | No | Yes |
| Power Consumption | 100 W | 75 W | 120 W |
| Price (Approximate) | $50,000 | $30,000 | $60,000 |
| Spectral Calibration | Factory calibrated | Requires calibration | Factory calibrated |
| Software Support | User-friendly GUI | Limited software | Comprehensive software |
| Application Examples | Agriculture, Geology | Environmental Studies | Remote Sensing |
| Data Processing Capabilities | Real-time processing | Limited processing | Advanced processing |
| Warranty and Support | 2 years | 1 year | 3 years |
| Availability | Widely available | Limited availability | Widely available |
| Weight | 5 kg | 3 kg | 6 kg |
| Dimensions (L x W x H) | 30 cm x 20 cm x 15 cm | 25 cm x 15 cm x 10 cm | 35 cm x 25 cm x 18 cm |
| Additional Features | GPS, IMU, Radiometric | None | GPS, IMU, Pan-Sharpening |
Types of Hyperspectral Sensors on Aircraft and Satellites
| Sensor Type | Key Characteristics | Applications | Platforms |
|---|---|---|---|
| Pushbroom | High spectral and spatial resolution, continuous data | Agriculture, mineral exploration, forestry | Aircraft, Satellites |
| Whiskbroom | High spectral resolution, lower spatial resolution | Oceanography, atmospheric studies | Satellites |
| Snapshot | Simultaneous capture of entire scene | Medical imaging, surveillance, astronomy | Satellites, Airborne Systems |
| Dual-Pushbroom | Simultaneous collection of two spectral ranges | Geology, coastal monitoring | Aircraft |
| Fourier Transform | Interferometric technique, high spectral resolution | Environmental monitoring, geology | Aircraft, Satellites |
| Prism | Moderate spatial and spectral resolution | Urban planning, vegetation analysis | Satellites |
| AISA (Airborne Imaging Spec) | High spectral resolution, variable configurations | Environmental monitoring, agriculture | Aircraft |
| CASI (Compact Airborne Spec) | Good spectral and spatial resolution, compact design | Precision agriculture, land cover mapping | Aircraft |
| HSI-3 | Hyperspectral and LiDAR fusion | Vegetation health assessment, topography | Airborne Systems |
| HySpex | High spectral resolution, versatile configurations | Mineral exploration, remote sensing | Aircraft, Satellites |
| EnMAP (Environmental Mapping) | Wide spectral range, medium resolution | Land use classification, forestry | Satellites |
| WorldView-3 | High spatial resolution, limited spectral range | Urban planning, disaster assessment | Satellites |
| Hyperion | High spectral resolution, limited spatial resolution | Geological mapping, environmental analysis | Satellites |
| Sentinel-2 | Moderate spatial and spectral resolution | Agriculture, land cover, disaster monitoring | Satellites |
| Landsat-8 | Moderate spatial and spectral resolution | Land use, environmental monitoring | Satellites |
Hyperspectral Remote Sensing Imagery (HRSI) Data Processing and Analyzing
Data PreProcessing
Hyperspectral data comes with its own sets of challenges, mainly due to its high dimensions and the complexity of the spectral data. Some of the key challenges are,
High Dimensionality
Hyperspectral images are characterized by their high dimensionality. With hundreds of spectral bands, it can be hard to process the dimension of hyperspectral data.
Missing Labeled Samples
Acquiring hyperspectral images can be relatively easy but accurately labeling the collected information can be really channel. This shortage of the labeled information can be a real challenge in classifying the hyperspectral images.
Spectral Variability Across Space:
The spectral information in hyperspectral images can be different across the special dimensions due to many factors such as atmospheric conditions, sensor variations and ground features. this difference can make the classification process complicated.
Image Quality
Image quality is an important factor in hyperspectral collection. Noise and errors during the data acquisition can significantly impact data quality affecting the accuracy of the classification process
To overcome these challenges, Hyperspector data undergoes a range of preprocessing steps. Unlike Multispectral and RGB sensors, the radiometric and atmospheric calibration workflow for hyperspectral data are more complicated.
Hyperspectral image preprocessing workflow
The workflow includes several essential steps:
Spatial Calibration
Spatial calibration is nothing but aligning each image pixel with its known units or features, providing information about spatial dimensions and correcting the optical distortions such as the smile and keystone effect. However, calibrating can be again misaligned by factors such as the changes in environment and the equipment.
Dimensionality Reduction
Since there are a high number of spectral bands in the hyperspectral images, dimensional reduction is really important. The process of converting high-dimensional data into a low-dimensional space while keeping the spectral information intact. Dimensionality reduction not only improves the processing speed but also increases classification accuracy.
Hyperspectral Image Classification
HyperSpectral Image Classification is important in extracting useful information from the hyperspectral data. It categorizes pixels in the image into predefined classes or clusters based on the spectral characteristics. The hyperspectral images are classified into 3 main types.
- Supervised
- Unsupervised
- Semi-Supervised
Supervised Classification
In supervised classification, labeled data is used for training. Common supervised classification methods include Support Vector Machines ( SVM ), Artificial Neural Networks ( ANN ), Decision Tree and Maximum Likelihood Classifications. These methods are used to determine discrete criteria based on known sample categories and prior knowledge.
Unsupervised Classification
Unsupervised classification, on other hand, doesn’t rely on prior labeled images. It categorizes pixels based on spectral similarities, often involving clustering without prior information. Methods like K-means Classification and Iterative Self-Organizing Data Analysis Technique ( ISODATA ) fall under this category
Semi-Supervised Classification
Semi Supervised Classification combines both labeled data and unlabeled data for training. This method compensates the limitations of both supervised and unsupervised learning methods. By using the labeled and unlabeled data in the feature space, semi-supervised classification enhances the classification accuracy of this method.
Potential Advancements in Hyperspectral Technology
The evolution of HyperSpectral sensor tech is an ongoing process. Researchers and engineers are continuously working on improving the sensor capabilities, including increases in the spatial and spectral resolutions. These advancements will help in acquiring more detailed and accurate data collection and enhance the effectiveness of the HyperSpectral Remote Sensing.
Conclusion
In conclusion, HyperSpectral Remote Sensing offers a unique capability to capture a wide range of narrow electromagnetic spectrum data, enabling in-depth analysis of Earth’s surface. The data is then organized into a hypercube, which can be used in information extraction. The evolution in the technology shows promising improvements in data collection and processing, thus enhancing the effectiveness of HyperSpectral Remote Sensing.
