Characterization and identification of weeds in the field using hyperspectral imaging
Carrying out weed characterisation and classification is crucial for weed removal and field management in smart agriculture,however,the problem of low accuracy of weed identification has not been explored in depth.In this study,a hyperspectral database of seven types of weed species in agricultural areas was established using field weeds as research objects.Firstly,hyperspectral images were collected using a ground-based hyperspectral camera and representative spectral curves,spectral profile features for each species were selected and analysed,and principal componet analysis was carried out for each species to reveal differences between weed spe-cies.The raw spectral data were then preprocessed using multiple scattering correction,normalisation,and first-order and second-order difference derivation.Finally,the support vector machine with a one dimensional convolution neural networks classification model was established and applied in identifying weeds in the hyperspectral images.The results suggested that the MSC-1DCNN model shows the best classification effect,with user accuracies ranging from 95.71%to 100%.This research not only provides strong hyperspectral characterisation of weed species and weed management,but also contributes to the development of automated weeding robots and the implementation of unmanned farms.
万方数据
维普
