The juvenile processing field tomato plant (Lycopersicon esculentum) was used as the representative crop plant. The commercial tomato fields in Yolo County, California were used as the experimental fields. The outdoor image formation system was designed to work in the real-time, field environment with all the attendant problems of travel speed, shaking of equipment, and variation in the position of camera. To capture high quality, uniformly illuminated images, factors such as sunlight conditions, shadow problems, optional light sources, diffusers, filters etc. were also investigated
A robust, environmentally adaptive, segmentation algorithm (EASA) was developed. This
is a partially supervised learning procedure used to overcome the major outdoor imaging
problems of light source temperature shifting, light source position changing, and
shadowing in the field of view. With EASA the operator needs only to choose the object
class(es) from the clustering result. Then the system is capable of building a
segmentation look-up table (LUT) in the field, based on prevailing environmental
conditions.
The results of this prototype system showed that a machine vision system
could be developed for outdoor, real-time field operation using current technology. This
research led to the conclusion for the first time that an outdoor field crop plant could
be successfully detected with semantic leaf-shape features and the center (stem) of the
plant could be located using the whole plant structural characteristics. A carefully
designed outdoor image formation system can simplify the algorithm for image segmentation
and crop plant pattern recognition. Thus, computationally expensive procedures are
avoided, making the real-time usage practical.
 |
“Instead of studying test plants in individual pots from a greenhouse, we
trained and tested our system in real-life, outdoor field condition.”
Go back to the Department Home Page