Module 1 The Remote Sensing Process

Introduction

Figure 1-1.1 How Remote Sensing Works

Digital image processing helps further this goal by allowing a scientist to manipulate and analyze the image data produced by these remote sensors in such a way as to reveal information that may not be immediately recognizable in the original form.

The Remote Sensing Process

Figure 1-1.2 The Remote Sensing Process

Identifying the Problem

The first step in remote sensing, as in any scientific study, is the definition of a problem.  Due to its multidisciplinary nature, the problems that remote sensing can be applied to are numerous and diverse.  In spite of this, the approaches to remote sensing can be categorized as being either scientific in nature or technological in nature.  The distinction is primarily a function of the motive behind solving the problem.  Scientific approaches are driven primarily by "curiosity or whim" (Curran, 1987) while technological approaches are driven by human need.  The methodology that is subsequently applied to the problem is usually dependent upon the origin of your problem.

There are three basic types of logic that can be applied to a problem; inductive, deductive, and technologic.  Scientific approaches use both inductive logic and deductive logic methodologies, while a technological approach uses a technologic logic methodology.  The steps in each of these logic methodologies can be seen in Figure 1-1.3.
 

Figure 1-1.3 Methodologies in Remote Sensing

Inductive logic could be described as learning logic.  The inductive methodology seeks to form tenable theories by making observations of phenomena, classifying these observations and making generalizations that form the basis of theories.  Most people use inductive logic every day.  For example, a person slips and falls on water which has spilled on their bathroom floor.  They would make the observation that when the tile in their bathroom gets wet, there is subsequent loss of traction.  This observation can then be generalized to a theory that all tile, when wet, provides less traction than when dry.

This type of logic is at the center of remote sensing when the focus is image interpretation.  Like our everyday learning experiences, a researcher using this logic observes facts about remotely sensed data and seeks to form general theories or principles that can be applied to other remotely sensed data (Curran, 1987).  Theories formed from this inductive approach are often fed directly into a deductive methodology (see Figure 1-1.3) where hypotheses are developed for testing the theories.

The focus of deductive logic is the formulation of theories and the subsequent testing of hypotheses.  Once a problem is identified, a researcher conjectures a theory to solve it.  To determine the validity of any such theory, hypotheses are developed and tested.  The hypotheses are at the core of the deductive logic.  Because of their importance, great care should be taken to formulate a hypothesis that is appropriate to the problem at hand.  Two of the most common types of hypotheses are the factual and the inferential.  A factual hypothesis clearly states a position that can be either verified or falsified.  (ex. There is a road that connects field A with field B.)  It is possible to verify this hypothesis as either truth or falsehood.  An inferential hypothesis is one which can be falsified.  Observations that fail to disprove the hypothesis do not necessarily prove its truthfulness.  However, a failure to disprove the hypothesis generally results in the acceptance of the theory being tested with the knowledge that future observations may later reverse that decision.

A technological approach differs from both the inductive and deductive in both its origin and its goal.  The basis of this approach is human need rather than scientific inquiry.  The goal is the rectification of that need rather than simply an increase in knowledge.  The focus of a technological methodology is the design of coherent plan which successfully blends "inputs from science, economics, aesthetics, law, logistics and other areas of human endeavor" (Curran, 1987).  Once a plan of action has been designed it is implemented without a formal hypothesis being stated.
 
 

Data Collection

In Situ Data

Remotely sensed data is being used in numerous fields and for a wide variety of applications.  Consequently, the collection of in situ data may take the form of field sampling, laboratory sampling, or some combination of both.  The techniques for these types of data collection should ideally be learned from the physical and natural science courses most related to the specific field of study such as chemistry, biology, forestry, soil science, hydrology, or meteorology.  When in situ data is to be used with remotely sensed data, it is important (for reasons explained elsewhere) that the positions of these data are known in relation to the remotely sensed data.  Due to ease of use and increasing affordability, global positioning system (GPS) receivers are the ideal tool to be used to gather such positional data when needed.  Using a GPS receiver, an x, y, and z coordinate can quickly be obtained to identify and locate individual samples in relation to remotely sensed data.

Remotely Sensed Data

Although most remote sensors collect their data using the basic principles described above, the format and quality of the resultant data varies widely.  These variations are dependent upon the resolution of the sensor.  There are four types of resolution that effect the quality and nature of the data a sensor collects: radiometric, spatial, spectral and temporal.  Radiometric resolution refers to the sensitivity of the sensor to incoming radiance (i.e., How much change in radiance must there be on the sensor before a change in recorded brightness value takes place?).  This sensitivity to different signal levels will determine the total number of values that can be generated by the sensor (Jensen, 1996).

Spatial resolution is a measurement of the minimum distance between two objects that will allow them to be differentiated from one another in an image (Sabins, 1978; Jensen, 1996).  This is a function of sensor altitude, detector size, focal size and system configuration.  For aerial photography the spatial resolution is usually measured in resolvable line pairs per millimeter on the image.  For other sensors it is given as the dimensions, in meters, of the ground area which falls within the instantaneous field of view of a single detector within an array - or pixel size (Logicon, 1997).  Figure 1-1.4 is a graphic representation showing the differences in spatial resolution among some well known sensors.
 

Figure 1-1.4 Comparison of Spatial Resolution of Common Sensors

Sensors also are unique with regard to what portions of the electromagnetic spectrum they see.  Different remote sensing instruments record different segments, or bands, of the electromagnetic spectrum. The number and size of the bands which can be recorded by a sensor determine the instrument's spectral resolution.  A sensor may be sensitive to a large portion of the electromagnetic spectrum but have poor spectral resolution if its sensitivity is contained in a small number of wide bands.  Another sensor that was sensitive to the same portion of the electromagnetic spectrum but had many small bands would have greater spectral resolution.  Like spatial resolution, the goal of finer spectral sampling is to enable the analyst, human or computer, to distinguish between scene elements.  More detailed information about the how individual elements in a scene reflect or emit electromagnetic energy increase the probability of finding unique characteristics for a given element, allowing it to be distinguished from other elements in the scene.  Figure 1-1.6 illustrates this principle by showing the spectral reflectance curves, or spectral signatures, generated when two sensors are used on the same target.  Both sensors cover the same range of the electromagnetic spectrum (2 to 2.5 um).  The solid bars at the top of the graph represent the specific segments of electromagnetic energy that each sensor can detect and record.  The first sensor (shown in red) has 17 bands in this range, while the second sensor (shown in blue) records the energy in only four bands.  As can be seen, the reflectance curve of the first sensor has greater detail, which may be useful in distinguishing its target from other objects with similar compositions.
 

Figure 1-1.6
Reflectance curves produced when a target is sensed using a sensor with a higher spectral resolution (Red) vs. a lower spectral resolution (Blue).

By increasing one or any combination of these resolutions, a scientist will increase the chance of obtaining remotely sensed data about a  target that contains accurate, realistic, and useful information.  The downside to increased resolution is the need for increased storage space, more powerful data-processing tools (hardware and software), and more highly trained individuals to perform or guide analysis (Jensen, 1996).  For these reasons, it is important to determine the minimum resolution requirements needed to accomplish a given task from the outset.  This will avoid time wasted unnecesarily processing more data than is needed.  It will also help to avoid the problem of too little data to allow completion of the task.
 

Data Analysis

Analog Image Processing

Table 1-1.1 shows the most commonly used elements of image interpretation used in visual image analysis.  The extent to which each of these elements is used depends on the not only on the area being studied, but the knowledge the analyst has of the study area.  For example, if an analyst has little or no knowledge of an area depicted in an image they may use the shape of objects to distinguish manmade objects from naturally occurring ones.  The texture of an object is also very useful in distinguishing objects that may appear the same if judging solely on tone (i.e., water and tree canopy may have the same mean brightness values, but their texture is much different (Schott, 1997)).  Association is also a very powerful image analysis tool when coupled with a general knowledge of the site.  For example, a building in the United States that has a fairly large parking lot and is near a circular track and football field is identified as most likely being a high school. 
 

 This is one of the areas of image processing that humans excel at - extracting information from images by combining multiple elements of image interpretation.  This is because we are continually processing images in our everyday life.  As we walk down the street we see the cars, other people, take note of the weather, etc.  All these images are passed to our brain where all of our experiences and learning are used to extract the most pertinent information.  Similarly, we are very adept at applying collateral data and personal knowledge to the task of image processing.  This, combined with the multi-concept of examining remotely sensed data in multiple bands of the electromagnetic spectrum (multispectral), on multiple dates (multitemporal), at multiple scales (multiscale) and in conjunction with other scientists (multidisciplinary), allow us to make a judgment not only as to what an object is, but its significance.  Other tasks performed in analog image processing include the optical photogrammetric techniques allowing for precise measurement of the height, width, location, etc. of an object (Jensen, 1996).  Many of these tasks are summarized in Figure 1-1.5.
 
 

Figure 1-1.5  Basic Steps in Analog and Digital Image Processing

Digital Image Processing

As the term implies, digital image processing is not only a step in the remote sensing process, but is itself a process which consists of several steps.  It is important to remember that the ultimate goal of this process is to extract information from an image that is not readily apparent or is not available in its original form.  The steps taken in processing an image will vary from image to image for multiple reasons, including the format and initial condition of the image, the information of interest (i.e., geologic formations vs. land cover), the composition of scene elements, and others which will be discussed throughout the modules in this volume.  There are three general steps in processing a digitial image; preprocessing, display and enhancement, and information extraction (Jensen, 1996).
 

Preprocessing - Before digital images can be analyzed, they usually require some degree of preprocessing.  This may involve radiometric corrections, which attempt to remove the effects of sensor errors and/or environmental factors.  A common method of determining what errors have been introduced into an image is by modeling the scene at the time of data acquisition using ancillary data collected .

Geometric corrections are also very common prior to any image analysis. If any type of area, direction or distance measurements are to be made using an image, it must be rectified if they are to be accurate.  Geometric rectification is a process by which points in an image are registered to corresponding points on a map or another image that has already been rectified.  The goal of geometric rectification is to put image elements in their proper planimetric (x and y) positions.
 

Information Enhancement -  There are numerous procedures that can be performed to enhance an image.  However, they can be classified into two major categories: point operations and local operations.  Point operations change the value of each individual pixel independent of all other pixel, while local operations change the value of individual pixels in the context of the values of neighboring pixels.  Common enhancements include image reduction, image magnification, transect extraction, contrast adjustments (linear and non-linear), band ratioing, spatial filtering, fourier transformations, principle components analysis, and texture transformations (Jensen, 1996).

Information Extraction - Unlike analog image processing, which uses all of the elements listed above (Table 1-1.1), digital image processing presently relies almost wholly on the primary elements of tone and color of image pixels.
There has been some success with expert systems and neural networks which attempt to enable the computer to mimic the ways in which humans interpret images.  Expert systems accomplish this through the compilation of a large database of human knowledge gained from analog image interpretation which the computer draws upon in its interpretations.  Nueral networks attempt to 'teach' the computer what decisions to make based upon a training data set.  Once it has 'learned' how to classify the training data succesfully, it is used to interpret and classify new data sets.
 

Information Output

Curran, P. J., 1987, "Remote Sensing Methodologies and Geography." International Journal of Remote Sensing, 8:1255-1275.

Drury, S. A., 1990, A Guide to Remote Sensing: Interpreting Images of the Earth. Oxford: Oxford University Press.

Jensen, J. R., 1996, Introductory Digital Image Processing: A remote sensing perspective, 2nd Edition. NJ: Prentice-Hall.

Logicon Geodynamics, Inc., 1997, Multispectral Imagery Reference Guide. VA: Logicon Geodynamics, Inc.

Lillesand, T. M., R. W. Kiefer, 1994, Remote Sensing and Image Interpretation, 3rd Edition. John Wiley & Sons, Inc., 23 p.

Sabins, F. F., 1987, Remote Sensing: Principles and Interpretation. NY: W.H. Freeman and Company.