Deep Learning in Hyperspectral Image Processing
Digital image data
Digital images are displayed using a collection of pixels (smallest building block of an image) to represent the spatial texture of an image. Each pixel can be associated with a number or an array of numbers, representing reflectance characteristics of objects at a specific range of spectral wavelengths.
For example, each pixels of an image, taken by a standard camera can be associated with an array of length three (for red, green, and blue channels), representing reflectance characteristics corresponding to visible range of spectral wavelengths (~ 400-700 nano meter). The image data captured by satellite sensors typically cover a much wider spectral wavelengths and each pixel is associated with an array of numbers containing tens of values depending on the spectral resolution of the sensor.
Digital image Processing
Digital image processing has gained a significant momentum with the advent of powerful computing hardware and measuring sensors as well as the advances in data management and development of data-driven modeling tools. Image data are used in a wide range of areas with analysis types including environmental modeling and monitoring, precision-agriculture, weather forecasting, medical imaging, mineralogy, etc.
Example
DataOrbs Inc. provides AI-enabled tools that are specialized for a wide range of image data processing and analyses. For example, one specific area of application is the application of hyperspectral image data in environmental system modeling in which image segmentation is required to characterize spatial heterogeneities in a watershed or agricultural system (see example of image segmentation using different deep learning algorithms).
In the context of environmental system modeling, DataOrbs Inc. has developed AI-enabled modeling tools for predicting land cover, soil conditions, and plant health characteristics using image data. In addition, DataOrbs Inc. has developed analysis tools that utilizes image data collected using satellite or air-born sensors for estimating model parameters of process-based watershed models like the Soil and Water Assessment Tool (SWAT).
