Weed Seeds Recognition Using Color PCA

In this paper, a novel approach for the recognition of weed seeds, known as color principal component analysis (PCA), is presented. This experiment involves two steps: dimensionality reduction with color PCA and classification. In dimensionality reduction part, color is used as an important element to identify weed seeds. To perform the recognition of color weed seeds images, we use the features of a 3D color tensor to generate vector spaces, and then color PCA is used to map the high dimensional spaces into low dimensional subspace and extract features. In classification part, minimum Euclidean distance classifier is used to recognize different weed seeds. Experimental results demonstrate that the representation of color PCA is superior significantly as expected and gains much higher recognition rates, amounting to 80.8%, which is 4.4% higher than those of traditional PCA.