Improvement of a Harvester Based, Multispectral, Seed Cotton Fiber Quality Sensor

ABSTRACT: An image-based multispectral sensor for in-situ seed cotton fiber quality measurement was developed and tested at Texas A&M University.  Results of initial testing of the sensing system with machine harvested seed cotton showed promise, with an R² value of .56 upon comparing the system's estimates to measured micronaire.  Since then improvements have been made to the system and measurement method in order to increase the accuracy of the sensor's micronaire estimates relative to HVI measurements.  The sensor takes six images of a sample of seed cotton, three images in near-infrared wavebands, and one image in each red, green and blue wavebands.  A color composite image is created to identify pixels in the image that are not cotton fiber so they can be excluded from the measurement, and the fiber quality is then determined with the remaining pixels from the NIR wavebands.  The image acquisition process has been automated, and efficiency of the image processing has been improved to allow this sensor to be capable of operation on a cotton harvester.  Fifty machine harvested seed cotton samples were collected from a 140-acre field at Texas A&M's IMPACT Center, and micronaire was estimated by manually presenting the samples to the sensor.  A portion of each sample was ginned, and the quality of the lint was measured with HVI.  A model was generated to relate estimates from the sensor to HVI measurements.  Results with the improved system will be reported in the manuscript.