A tool to overcome the challenges of green plant segmentation in hyperspectral images

Dr Huajian Liu, from the APPF’s Adelaide node, has developed a one-class support vector machine classifier combined with a pre-processing method named ‘hyper-hue’ to segment green plant pixels in hyperspectral images.

Green plant segmentation plays an important role in hyperspectral-based plant phenotyping, but existing image segmentation methods are dependent on data types, plants and backgrounds, and might not utilise the power of hyperspectral data.

"Experimental results showed that our method out-performed the approaches using vegetation indices or SVM only", said Dr Liu.

The model was trained using the data of wheat and worked equally well for other species, and the modelling method was suitable for both VNIR and SWIR data.

"In the future, this green plant segmentation method will be further tested using data collected in the field, such as on aircraft or ground-based vehicles", added Dr Liu.

The hyper-hue algorithm is free to download here or here.

Read a background paper on the hyper-hue algorithm here.

Contact Dr Liu here.

11 September 2019