| Two novel non-morphological machine vision algorithms that are robust against occlusion were developed to differentiate crop plants from weeds in real-time. The first of these two methods was designed to distinguish Acala cotton varieties from weeds using the presence of red pigment in the cotton leaf tissue where the petiole attaches to the leaf (Red Dot method). The second method was designed to differentiate dicots from monocots through the use of a real-time erosion technique (Erode method).; In validation tests with 141 images of cotton combined with nightshade (Solanum nigrum), and nutsedge (Cyperus esculentus L.) weeds, the Red Dot algorithm identified and marked for spray 76% of the weeds, while correctly identifying and not marking for spray 80% of the cotton plants. The Erode algorithm was evaluated using 326 images of cotton, sunflower, and bean crops with their associated weed species. The Erode algorithm had a successful species total classification rate of 93%, 92% 94%, and 79% in sunflower-morning glory, bean-morning glory, cotton-grass, and cotton-nightshade image sets respectively.; A real-time robotic weed control system was developed and tested utilizing the erode algorithm in commercial cotton fields in the San Joaquin Valley from April through May 1999. The weed control system was designed to distinguish weeds from cotton plants, and applied a chemical spray only on targeted weeds, while traveling at a continuous speed of 0.97 m/s. The robot consisted of a real-time machine vision system, a controlled illumination chamber, and a precision chemical applicator. The precision chemical applicator could spot spray targets with a resolution of 0.97 cm x 1.27 cm. In commercial cotton field tests the robot correctly sprayed 88.8% of the weeds, while correctly identifying and not spraying 78.7% of the cotton plants for a total classification success rate of 86.9%. The results of this system are better than any previously tested robotic weed control system as well as hand hoeing crews, which usually leave between 25--35% of the weeds remaining in the field. |
