Classification of broadleaf weed images using Gabor wavelets and Lie group structure of region covariance on Riemannian manifolds

A new method based on Gabor wavelets (GW) and Lie group structure of region covariance (LRC) representation was applied to classification of broadleaf weed images on Riemannian manifolds. A total of 320 images of four different varieties of broadleaf weeds were used for analysis. The classification tasks were more difficult and challenging than previous tasks, because the weeds chosen had very similar texture characteristics. The optimal multi-resolution GWs were used to decompose the image into texture features and the LRC was used to extract the filtered image features on Riemannian manifolds. A new k nearest neighbour (KNN) classifier was presented for feature matching on Riemannian manifolds. Leave-one-out cross-validation was used to test the robustness of the classification model. The overall recognition accuracy was 93.13% and the average recognition time was 1.5 s using Matlab. The recognition accuracy and performance were superior to previously associated studies. The results demonstrate the potential of the proposed methods for real-time application to weed recognition.

► The optimal multi-resolution Gabor wavelets were used to decompose the weed image into texture features. ► Lie group structure of region covariance was used to extract the texture features on Riemannian manifolds. ► k nearest neighbour (KNN) classifier was presented for feature matching on Riemannian manifolds. ► The overall recognition accuracy was 93.13%.