Weed detection based on the optimized segmentation line of crop and weed

Weed detection is a key problem of spot spraying that could reduce the herbicide usage. Spectral information of plants is very useful to detect weeds in real-time for the fast response time. However, the cost of an imaging spectrograph-based weed detection system is too high. Therefore, the main objective of this study was to explore a method to classify crop and weed plants using the spectral information in the visible light captured by a CCD camera. One approach to weed classification was to directly use of G and R component of RGB color space. Another was to utilize the spectral information among the green band that hue was regarded as wavelength, and saturation was represented as reflectance. The result of statistic analysis showed that both of them using the G-R and H-S optimized segmentation line of crop and weeds could be used to detect weed (lixweed tansymnustard) from wheat fields. Moreover, the method of using the H-S optimized model could avoid the affect of lighting.