关键词: K均值聚类, 谷子冠层分割, H分量

Abstract: This research takes millet as the research object, collects four types of canopy images of millet, cloudy sky, complex background with shadows, uneven illumination, and dew and rain reflection. The extra-green segmentation, K-means clustering segmentation in Lab space and H-component of K-means clustering segmentation are used for canopy extraction, and the optimal method for millet canopy extraction under different conditions is explored. For millet canopy images with cloudy and complex backgrounds and shadows, the three algorithms can extract relatively completely, and the segmentation accuracy was above 93%; for images with uneven lighting, the ultra-green segmentation effect is the worst, based on the Lab space and the K-means clustering segmentation effect of the H-component is relatively excellent, respectively, 93% and 96%; for the image of dew and rain reflection, the K-means clustering segmentation accuracy based on the H component is the highest, reaching 97%. The results show that the K-means clustering algorithm based on H component is ideal for segmentation of millet canopy images obtained under four different environmental conditions, which provides a certain reference value for subsequent automatic monitoring of millet growth.

Key words: K-means, millet canopy segmentation, H-component