Our thematic products mostly derive from multi-spectral imagery using an approach that includes parametric multi-variate statistical methods combined with image interpretation expertise to simply assign thematic class values to image pixels.

Here are some classification terms . . .

• crisp - binary yes/no pixel assignment,
• fuzzy - probability-per-class pixel assignment,
• hybrid - combined with ancillary data,
• cart - classification and regression tree analysis.

Here are some classification methods and applications that we can do . . . 

Object Based Classification  -  We can segment imagery into homologous vector polygon objects by using scale, color, and texture parameters with hierarchical rules for per-pixel-per-polygon classification.

Pixel Based Classification  -  We can use spectral clustering / ISODATA (Iterative Self Organizing Data Analysis Technique) to implement various statistically based classifiers that consider each pixel's multi-layer value without consideration of its neighbors - (with innovative classifier supervision).

Classification Accuracy Assessment  -  Accuracy specifications cited in our proposals and provided in our products are meant to endure rigorous comparison against independent accuracy assessment surveys reporting Omission and Commission Error on a per-theme and overall basis.

Raster to Vector Feature Extraction  -  We can offer geometrically smooth lightweight accurate thematic data in topologically clean "line" and "polygon" formats from the following data sources.

• Multi-spectral and Panchromatic Imagery
• Lidar Data
• Paper Maps

Land Cover / Land Use (LC/LU) Change Analysis
Imagery based thematic change detection analysis provides methodological versatility when updating LC/LU layers.  A good change analysis product is augmented with a thematic "from-to" category layer and various statistics.  We are experienced with these general methods ...

• multi-temporal compositing  -  offers quick and potentially effective results.  It requires cloud-free imagery collected on near-anniversary dates by similar sensor models.
• cross correlation  -  offers more flexibility with source data requirements by allowing the initial ("from") image data to be thematically classified.
• post classification  -  offers the most flexibility by accepting thematically classified ("from") and ("to") input imagery.  It can be effectively used to mitigate extreme variation between ("from") and ("to") environmental conditions.