|LI, PAN - Northwest Agricultural & Forestry University|
|HE, DONGJIAN - Northwest Agricultural & Forestry University|
|QIAO, YONGLIANG - Northwest Agricultural & Forestry University|
Submitted to: ASABE Annual International Meeting
Publication Type: Proceedings
Publication Acceptance Date: 7/26/2013
Publication Date: 7/26/2013
Citation: Li, P., He, D., Qiao, Y., Yang, C. 2013. An application of soft sets in weed identification. ASABE Annual International Meeting. Paper No. 131620222.
Interpretive Summary: Herbicide applications effectively reduce weed infestations in crop fields but are only needed in infested sections of fields. Further, the weed species determine which type of herbicide will be most effective. In order to achieve effective weed management with minimum necessary application of herbicide, new methods are needed to remotely distinguish between various weed species and crops. In this study, weed species were classified from images of weeds and corn by the use of ‘soft set theory’. Soft set-based weed identification achieved 96% accuracy. The soft set-based weed identification provides a useful tool for weed management and reduced herbicide application.
Technical Abstract: Soft set theory is originally proposed as a general mathematical tool for dealing with uncertainties present in most of our real life. This study applied soft sets to improve low accuracy of weed identification caused by similar features. Firstly, three types of plant leaf features including shape, texture and fractal dimension were extracted from the plant leaves after a series of image processing. Then the weed-classification matrix went through arithmetic operations on the relation-matrices constructed with eigenvalues and their weight factor coefficients, from which the label corresponding to its largest membership in every row was selected. Finally the weed was distinguished according to the label. Also the soft set theory showed higher performance in terms of robustness and algorithm complexity comparing with the Bayesian classifier, support vector machine (SVM) and back-propagation (BP) neural network. The proposed method provides a useful tool for weed identification and selectively spraying herbicide.