A novel algorithm is proposed to identify weed infection area in an agronomic crop field by decomposing a digital image on a wavelet orthonormal basis. Specifically, the image in the near-infrared (NIR) spectrum is directly transformed from the spatial domain into the spatial-frequency domain by resorting to the fast pyramid wavelet transform without pre-segmenting the image or using any memory mapping techniques. The algorithm extracts spatial-frequency information to detect the weed-infection areas featured by high frequency components. The mathematical framework is derived to elucidate the proposed algorithm. The proposed algorithm is capable of not only identifying the weed-infection areas but also potentially determining the weed densities in the identified areas. The efficacy of the algorithm is demonstrated by analyzing images of corn and soybean fields. Moreover, the simplicity and low hardware requirements with the proposed algorithm make it feasible to be applicable to a weed-control system in real time. In fact, a prototype of sprayer system based on the resultant algorithm has been constructed and being tested in Department of Agricultural Engineering, University of Illinois.
|Original language||English (US)|
|Journal||Paper - American Society of Agricultural Engineers|
|State||Published - Dec 1 1997|
|Event||Proceedings of the 1997 ASAE Annual International Meeting. Part 1 (of 3) - Minneapolis, MN, USA|
Duration: Aug 10 1997 → Aug 14 1997
ASJC Scopus subject areas
- Agricultural and Biological Sciences (miscellaneous)