|LUO, BIN - Grenoble Institute Of Technology|
|CHANUSSOT, JOCELYN - Grenoble Institute Of Technology|
Submitted to: IEEE IGARSS Annual Proceedings
Publication Type: Proceedings
Publication Acceptance Date: 8/25/2011
Publication Date: 9/15/2011
Citation: Luo, B., Yang, C., Chanussot, J. 2011. Linear unmixing of multidate hyperspectral imagery for crop yield estimation. IEEE IGARSS Annual Proceedings. 2011 CDROM.
Interpretive Summary: Remotely-sensed hyperspectral imagery can be useful for estimating crop yield and its spatial variation in precision agriculture. This paper evaluated a new spectral technique, unsupervised linear unmixing, to estimate crop yield from hyperspectral images. The spectral scheme was used to convert the hyperspectral images taken from a grain sorghum field on two dates to multiple crop cover abundance maps. Statistical analysis showed crop yield was significantly related to the crop abundance maps. By combining the vegetation abundances extracted on the two dates, the correlations with yield are significantly improved. These results indicate that unsupervised linear unmixing is a useful tool to convert hyperspectral imagery of crop fields to relative yield maps.
Technical Abstract: In this paper, we have evaluated an unsupervised unmixing approach, vertex component analysis (VCA), for the application of crop yield estimation. The results show that abundance maps of the vegetation extracted by the approach are strongly correlated to the yield data (the correlation coefficients are from 0.7 to 0.8). The results validate the higher efficiency of the unsupervised unmixing approach compared with the supervised methods used in previous studies for the purpose of yield estimation. In addition, the unmixing was performed on the hyperspectral images taken on two different dates. The results show that the correlations between the vegetation abundances and the yield image might change in terms of the observation dates. However, by combining the vegetation abundances extracted on different dates (by computing their products or the square roots of the products), the correlations are significantly improved.