Weed detection in multi-spectral images of cotton fields

Victor Alchanatis*, Leonid Ridel, Amots Hetzroni, Leonid Yaroslavsky

*Corresponding author for this work

Research output: Contribution to journalArticlepeer-review

71 Scopus citations


A means for automatic detection and evaluation of weeds in the field was developed and tested; it was based on an acousto-optic tunable hyperspectral sensor and a detection algorithm. The algorithm that was developed used spectral reflectance properties and robust statistics features for weed detection. Soil-crop segmentation was done with two spectral channels, chosen from 100 channels available from the hyperspectral sensor. Weed detection was based on texture features, extracted from the segmented images. The algorithm was applied to a database of images of cotton plants and weeds, in their early stages of development. The results showed a good detection ability. The occurrence of weeds was detected in all images; the weed-infested area was estimated with 14% error, and the false detection rate was 15%.

Original languageEnglish
Pages (from-to)243-260
Number of pages18
JournalComputers and Electronics in Agriculture
Issue number3
StatePublished - Jun 2005


FundersFunder number
United States - Israel Binational Agricultural Research and Development Fund


    • Acousto-optic tunable filter
    • Classification
    • Cotton
    • Detection
    • Spectral reflectance
    • Weeds


    Dive into the research topics of 'Weed detection in multi-spectral images of cotton fields'. Together they form a unique fingerprint.

    Cite this