Submitted to: National Cotton Council Beltwide Cotton Conference
Publication Type: Proceedings
Publication Acceptance Date: 2/15/2009
Publication Date: 5/15/2009
Citation: Siddaiah, M., Whitelock, D.P., Lieberman, M.A., Hughs, S.E., Grantham, S. 2009. Categorization of extraneous matter in cotton using machine vision systems. National Cotton Council Beltwide Cotton Conference. 2009 CD 1211-1216. Interpretive Summary: US Cotton, sold at home and on the world market, is classified by the Agricultural Marketing Service (AMS). This classification, used to determine the price of cotton, is based mostly on High Volume Instrument machine measurements, except leaf grade and extraneous matter are determined by a somewhat more subjective human visual classification and can be time consuming and labor intensive. A machine based measurement of extraneous matter may aid the human classer in consistently making extraneous matter calls. A Cotton Trash Identification System (CTIS) developed at the USDA-ARS Southwestern Cotton Ginning Research Laboratory in Mesilla Park, New Mexico was evaluated by comparing it to AMS classer extraneous matter calls for over 800 samples collected by the USDA-AMS Cotton Program, Standardization & Engineering Branch in Memphis, Tennessee. There was good agreement by CTIS to the classer in the identifying extraneous matter, although CTIS tended to identify extraneous matter more often than the classer. Future work will focus on strengthening the agreement of CTIS to the human classer and may lead to an aid for the classer to help make more reliable and accurate cotton extraneous matter calls.
Technical Abstract: The Cotton Trash Identification System (CTIS) developed at the Southwestern Cotton Ginning Research Laboratory was evaluated for identification and categorization of extraneous matter in cotton. The CTIS bark/grass categorization was evaluated with USDA-Agricultural Marketing Service (AMS) extraneous matter calls assigned by human classers for 210 cotton bale samples. AMS classers assigned extraneous matter calls on four cotton faces of a sample from a given bale of cotton. Scanner acquired images of the same four faces at 400 DPI and 800 DPI resolutions were analyzed to evaluate the CTIS performance. Soft computing techniques were used to identify trash objects in the acquired cotton images (4 in. x 7 in.) and categorize the objects into bark/grass, stick, leaf, and pepper trash categories. The primary goal of the study was to evaluated and calibrate CTIS categorization of extraneous matter using classer extraneous matter calls. CTIS agreed with the classer call 97% of the time when there was a classer extraneous matter call and 43% of the time when there was no classer call. CTIS may find a place as a tool to aid human classers in the classification of cotton by helping to identify extraneous matter.