GIS 4035 - Module 09: Unsupervised Image Classification

This week we dealt with unsupervised image classification. This is the process of using an algorithm to differentiate raster image pixels (often across and comparing multiple bands of an image set) in order to create classifications that produce (It is hoped) classifications of pixels that reflect actual ground features. The identification of groupings (and adjustment of classifications) are done after the classification is processed. This stands in contrast with supervised classification where an algorithmically based system is trained before classification by receiving input on areas across the image space that are grouped together into classifications.

The map below was produced by running an unsupervised classification on an orthophotograph of UWF's main campus. The initial classification was set to produce 50 classes. Once those were created, they were each identified as belonging to one of five very simple classes: Trees, grass, building/road, shadow or mixed. The mixed class was reserved for classes that were ambiguous as to their proper identification - encompassing a mix of grass and road. Once all the classes were identified, the classes were collapsed to encompass just those five and the map was created.

UWF Campus Thematic Raster Map

GIS 4035: Module 8 - Thermal Imagery

This week's lab involved manipulating multispectral imagery and using bands in the thermal infrared range to do remote sensing work. While the map below does not display imagery in the thermal infrared (Band 6 in imagery provided by the Enhanced Thematic Mapper Plus suite of satellites), it was used as part of the map analysis performed this week. The map identifies a field network of rice paddies near Guayaquil, Ecuador. Their appearance is somewhat anomalous in the Near Infrared. Healthy vegetation tends to be highly reflective and appear bright in Near Infrared bands (bands 4, 5 and 7 in ETM+), but these fields are dark in those ranges, while appearing like normal, green fields in natural color (where the visible red, green and blue bands are displayed as red, green and blue). In the Thermal Infrared, where captured EMR tends to be the result of heat emissivity rather than reflected, these fields were also intensely dark. During the day (when these images were recorded), water tends to be the coolest material recorded. Vegetation is usually cool, too, but is still somewhat warmer than water. The field'd Thermal IR appearance matched nearby water bodies closely and were darker than areas of healthy vegetation. The inference made was that these were rice paddies - flooded fields planted with rice seedlings. In the NIR and TIR, this gave the fields the appearance of being water, but the seedlings managed to show the signs of photosynthesis to the visible bands (absorbing blue and red, reflecting green to some degree). The latter is probably due to the shallowness of the water.

Rice Paddies in Guayaquil, Ecuador (shown in 'natural' false color)

GIS 4035: Module 7 - Image Preprocessing 2: Spectral Enhancement and Band Indices

This week continued the work with Erdas Imagine and satellite imagery, beginning a deeper delve. Imagery from Landsat 5 Thematic Mapper was provided with the six bands that maintain the same spatial resolution included (Blue, Green, Red, Near Infaread and two mid-level Infrared bands). Several exercises provided practice manipulating and interrogating the images, combining various bands so as to make various features stand out. For example, the most common false color combination where the display layers for Red, Green and Blue are re-mapped respectively to Near IR, Red and Green was given a good workout.

The deliverables this week were created as part of a treasure hunt. The features were described spectrally and had to be located somewhere in the study area of the imagery. Feature 1 represented typical water features, which tend to absorb most electromagnetic radiation in the near infrared band. Feature 2 turned out to be snow, ice and glaciers, which appear bright (high radiometric values) in bands 1-4, but quite dark in bands 5 and 6. Feature 3 was an unusual water feature that had radiometric gradations in 1 - 4 and was somewhat brighter than other water features in these bands. Each feature was displayed in a map with a different combination of bands being used along the RGB display bands.

Feature 1: A Lake

Feature 2: Ice & Snow

Feature 3: Another Lake, with radiometric gradations

GIS 4035: Module 6 - "Spatial Enhancement"

This week continued our introduction to Erdas Imagine with some practice using image filters on raster imagery. This included running various high and low pass filters on an image, as well as edge enhancing and other filters. The map below was produced as an attempt to rectify data errors produced on images from Landsat 7, which sometimes produces bands across an image without any data. A fourier transform was used to reduce the banding somewhat, and then a sharpening filter was run on the resulting image in order to improve the image resolution. The results were not great, but the banding was somewhat reduced compared to the original image.

Moderately corrected banding of Landsat 7 image

GIS 4035: Module 5a - Introduction to ERDAS Imagine

This week's assignment was a basic introduction to ERDAS Imagine, a software package designed for doing remote sensing work, specifically with the editing of raster images in mind. While going through the ins and outs of uploading various rasters, a subset of a larger Land Cover raster was exported from Imagine (after adding an area attribute within the software) which was then used to produce the map below in ArcGIS (thus avoiding an avowed bug in Imagine that would have crashed the system when attempting similar output).

Land Cover Map of a Random Portion of northwest Washington state.

GIS4035: Module 4 - Ground Truthing and Accuracy Assessment

This week's assignment built upon last week's. The Land Use Land Cover assessment that was compiled in Module 3 was assessed in this module. Since actual fieldwork is impracticable for an online course, Google Street Map was the next best thing. Thirty random points were generated in the study area, weighted by land classification so that each class used in the classification received at least one point and proportionally more points were assigned to classes with larger coverage areas. Each point was then located in google maps and the accuracy of the classification at that point itself was assessed and recorded. The map below was then compiled, showing the accuracy result for each point.

The accuracy calculated as a ratio of accurate point classifications to the total number of points assessed is 63.3%.

Accuracy Assessment for Pascagoula, MS Land Use / Land Cover Map

GIS4035: Module 3 - Land Use and Land Cover Classification

This week's assignment was to create a land use and land cover classification for an aerial image that was provided. The study area is a portion of Pascagoula, MS including a good portion of wetlands and open water. We were provided with a basic classification scheme based on the one used by the US Geological Survey and had to identify various land types and create polygons to mark off the various classes as we perceived them on the aerial image. Care had to be taken to keep all the classes at the same scale.

Land Use & Land Cover Map for Pascagoula, MS

GIS4035: Module 2 - Introduction to Visual Interpretation.

This week's assignment acts as a basic introduction to performing visual interpretation of aerial imagery. Two digitized aerial photographs were provided which were subjected to some basic interpretations. The first image was used to select areas based on tone from very light to very dark and then also based on texture from very smooth to very coarse. The second image was used to identify features in the image based on various attributes, such as the shape and size of an object, and object's shadow, patterns and associations.

Tone and Texture Identification

Feature Identification