This paper discusses a framework for implementation of a machine-vision system for on-line identification of trash objects commonly found in cotton. Soft computing techniques such as neural networks and fuzzy inference systems can classify trash objects into individual categories such as bark, stick, leaf, and pepper trash types with great accuracies. This identification of trash objects to individual categories can be used for the dynamic allocation of trash extraction equipment during the ginning process. Such a system can be implemented in a modern gin, to configure an optimal set of equipment during ginning to produce quality cotton. Classification of cotton in real-time allows for an automated means to assign trash grades to cotton, and could have a significant impact on the entire cotton industry.