2225 High Speed Machine Vision Sensing of Cotton Lint Trash

As machine design in the cotton ginning industry advances, the trend is towards systems that allow for dynamic adjustment of cleaning capabilities. There is however, a lack of real-time sensors suitable for detection of the trash content that would allow for determination of the trash content before the machine performs the cleaning. Current state of the art sensors typically sense the product’s trash levels after the cleaning has already taken place or so far before the cleaning takes place, it’s impossible to synchronize the cleaning to the lint being cleaning. This adhoc trash level sensing dictates that a large safety buffer is utilized to avoid under-cleaning the cotton in an effort to avoid commodity discounts that occur when the lint is under-cleaned. This research estimates improvement in efficiencies of upwards of 30% are possible, if the trash content was known a-priori via a real-time sensor that adjusts the machine to the actual lint being cleaned, as this would allow for a significant reduction in the dead-band safety buffer. One of the main hurdles to sensing the trash content dynamically is that current image processing systems are too slow to provide the information in time to adjust the industry’s new adjustable machines. This research examines the development of a new signal processing algorithms suitable for use in massively parallel vector processors. Performance tests of the new filter showed a processing speed gain in excess of 20 times faster resulting in a system that is capable of processing full 1024x1024 pixel image in less than 17 ms. At this speed, the image processing system’s performance is now sufficient to provide a system that would be capable of real-time feed-back control that is in tight cooperation with the cleaning equipment.