Skip to main content
ARS Home » Research » Publications at this Location » Publication #93183
Title: SPECTRAL REFLECTANCE PATTERN RECOGNITION FOR SEGMENTING CORN PLANTS AND WEEDS

Author
item HUMMEL, JOHN
item YU, JING - UNIV OF ILLINOIS

Submitted to: International Conference on Precision Agriculture Abstracts & Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 7/23/1998
Publication Date: N/A
Citation: N/A

Interpretive Summary: Commercially available spectral reflectance sensors can detect the presence of plant material against a background of bare soil. These systems are able to identify areas of weed infestation in fallow fields or in the area between the rows of a row crop and control the application of herbicides. In typical U.S. corn production, a number of options are available for controlling weeds in between the rows-e.g., broadcast herbicide application or mechanical cultivation. In-row weed control might be possible with non-selective herbicides if detected plant material areas can be segmented into crop and weed subsets. Accomplishing this task in real time would allow application of the herbicide to only the weeds, which could significantly reduce the amount of herbicide used. The results should be of value to scientists researching sensors for weed detection.

Technical Abstract: This study reports on the use of a sensor based on differences in spectral reflectance at the visible and near infrared wavelengths exhibited by chlorophyll-bearing plant tissue and soil to discriminate weeds from crop plants. The system uses solid-state, frequency-modulated light emitters for discriminating plants from soil. The sensor was incorporated into an eight channel device called the Patchen WeedScanner (Patchen California, Los Gatos, CA) with each channel equipped with a solenoid-valve-controlled spray nozzle. We used the industry-developed spectral reflectance sensor to identify and locate plant material, and developed experimental C- language algorithms to segment the corn plants and weeds. Four typical corn plant patterns were developed and used to test the segmentation efficiency of the algorithms. Tests were completed showing that the algorithms can, for specific corn plant sizes and weed sizes and densities, identify corn plants and distinguish them from weeds; and one algorithm in conjunction with the Patchen Weedscanner system was able to successfully identify corn plants with an accuracy of at least 75%.