| dc.description.abstract | For precision agriculture to work, an automated process to collect spatial-variability data within a field is necessary. Otherwise, data collection is prohibitively expensive and time consuming. Furthermore, to minimize measurement error due to harvesting method, data-collection processes involving normal cotton harvesting and ginning operations must be used.
For the case of cotton, an automated prototype system using image processing to measure the micronaire value of cotton fiber during harvest was designed and built in the laboratory. This system was tested with two image-processing algorithms to identify and remove the effects of objects present in the images that were not cotton fiber, and then measure the reflectivity in three Near-Infrared (NIR) wavebands. Both algorithms yielded similar results when used on seed cotton samples. The reflectivity measurement after removing the effects of foreign matter had a strong relationship to standard micronaire measurements (R^2= 0.73 and 0.74 for the ratio-image and single-image algorithms, respectively) with a root mean squared error (RMSE) of 0.28 and 0.27, respectively. The ratio-image pixel classification method classified an average of 58% of the pixels in an image as "cotton", while the single-image method classified an average of 81% of the pixels in each image as cotton. These results do not show as strong a relationship between micronaire and NIR reflectivity of cotton samples as previous research done with very uniform lint cotton calibration samples. This is attributed to the higher content of foreign matter in seed cotton samples. With higher trash cotton and fiber that has not yet been cleaned, results obviously are not as good as when using calibration cotton samples. These results indicate the system can be adapted to perform in-situ measurement of cotton fiber quality, specifically micronaire, and enable harvesters to create quality maps of a field automatically to allow better crop management. | en |
