🚀 go-pugleaf

Hi:

I am faced with following problem to be solved using image processing
techniques:

We are trying to classify different non-lint particles in a cotton sample
using imaging techniques.

Cotton sample after ginning contains different kind of trash material on the
basic of which cotton is graded for it quality. The trash (non-lint particles)
are of type pieces of leaf, bark, sticks, pepper (fine pieces of leaf), 
cotton seed coat fragments, etc. We are looking at different features
of these particles, area, perimeter, shape factor, eccentricity, moments etc.
Using these features we have achieved fairly good classification. We also 
used clustering approaches, Neural nets etc. 

Those who have seen some of these non-lint particles, is these anything else
that we need to consider as feature, in order to get a better 
classification ? Like most of real life problems, one major problem is 
of overlapping clusters of these trash types and increasing feature 
space may improve the classification.

Any suggestions and ideas are appreciated.

Regards

raj
USDA, ARS