Parallel Algorithm for GPU Processing; for use in High Speed Machine Vision Sensing of Cotton Lint Trash.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Textile Manufacturing

Background:

• Current cotton lint cleaning systems lack real-time trash sensing, leading to over-cleaning and significant lint loss.
• Existing machine vision systems face processing time limitations, hindering effective real-time control.
• Large safety buffers in cleaning machinery result in fiber damage and reduced yield.

Purpose of the Study:

• To develop a rapid trash sensing system for cotton lint cleaning.
• To improve processing times for machine vision systems in textile applications.
• To enable real-time feedback control for optimized cotton cleaning processes.

Main Methods:

• Utilized programmable graphic processing units (GPU) as an alternative computation platform.
• Developed a new parallel algorithm for cotton trash sensing and image processing.
• Implemented and tested the algorithm on a GPU for enhanced processing speed.

Main Results:

• Achieved a speed-up of over 6.5 times compared to optimized code on a central processing unit (CPU).
• Processed a 1024x1024 image in less than 17ms, enabling real-time capabilities.
• Demonstrated potential for up to 30% reduction in lint loss through tightly coupled sensing and control.

Conclusions:

• GPU-based processing significantly accelerates machine vision for cotton trash detection.
• The developed parallel algorithm and GPU implementation are suitable for real-time feedback control in cotton cleaning.
• This advancement promises to reduce waste, minimize fiber damage, and improve the efficiency of cotton processing.