Identification of gramineous grass seeds using Gabor and locality preserving projections

Forage identification is primarily realized by human experts at a low efficiency, which does not meet the requirements of a digital grassland management. In this study, we propose an automatic identification system for gramineous grass seed, an important category of forage in grassland, based on seed images using Gabor filters and local preserving projections (LPP). The system includes four modules: image acquisition, image preprocessing, feature extraction, and feature matching. Seed images are first captured by a common digital camera, and then preprocessed by a morphological operation to obtain the ROI. In the feature extraction module, the integration of Gabor filters and LPP can provide robust features for varying brightness and image contrast while preserving the manifold structure of the images for efficient dimensionality reduction. The nearest neighbor classifier and linear discriminant analysis (LDA) classifier are used for classification. The novelty of the system lies in two aspects; one is that gramineous grass seeds in the study is automatically identified as valuable resources in grassland, instead of the certain species of weed to be distinguished from crops in the previous weed classification. The other is that Gabor filter and LPP are applied to extract the textural manifold features for the identification of gramineous grass, rather than the geometric features of appearance of gray-scale images, for more robust performance. The experimental results demonstrate the effectiveness of the seed identification system.