研究目的
To segment shadow, soil, and vegetation in UAV-based RGB images of a vineyard using the Triangular Greenness Index (TGI) combined with k-means and CLARA clustering algorithms, aiming to improve remote crop monitoring techniques.
研究成果
The proposed methodology successfully segments soil, shadow, and vegetation in UAV-based RGB images using TGI and clustering algorithms. CLARA outperforms k-means in accurately identifying vegetation areas, reducing misclassification. This approach supports the development of advanced monitoring techniques for precision agriculture, such as water use efficiency estimation, by enabling precise separation of vegetative and non-vegetative materials.
研究不足
The study relies on visual analysis for performance comparison, which may be subjective. The algorithms are sensitive to outliers and lighting conditions, and the methodology is specific to RGB images and vineyards, potentially limiting generalizability to other crops or sensor types.
1：Experimental Design and Method Selection:
The study used a combination of the Triangular Greenness Index (TGI) derived from RGB images and two clustering algorithms (k-means and CLARA) for image segmentation. The rationale was to differentiate pixels representing vegetation, soil, and shadow in agricultural images.
2：Sample Selection and Data Sources:
Images were acquired from a commercial drip-irrigated vineyard in Pencahue, Maule Region, Chile, during the 2017-2018 season.
3：List of Experimental Equipment and Materials:
A DJI Phantom 3 Advanced UAV with a
4：76-megapixel RGB sensor was used for image acquisition. Software included Pix4D for image mosaicking and R Studio with the 'cluster' library for index processing and algorithm implementation. A personal computer (laptop) with an Intel Core I7 processor at 6 GHz and 16GB RAM running Windows 10 was used for computations. Experimental Procedures and Operational Workflow:
The UAV was pre-programmed for a flight at 30 meters altitude with 90% image overlap. RGB images were captured, and TGI was computed for each pixel. The k-means and CLARA algorithms were applied to segment the images into three classes (soil, shadow, vegetation), with visual analysis for comparison.
5：Data Analysis Methods:
Segmentation results were analyzed visually to assess performance, particularly in differentiating vegetation from other classes.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
