研究目的
To develop an efficient method for extracting the leaf region and counting the number of leaves in digital plant images to support plant phenotyping and agriculture.
研究成果
The proposed method achieves high accuracy in leaf segmentation (95.4% FBD) and counting (DiC of -0.7 and |DiC| of 2.3), outperforming existing state-of-the-art methods. It is robust, requires no training, and is applicable across various phenotyping platforms with minimal modifications. Future work includes making it open-source for broader use in plant phenotyping.
研究不足
The method is primarily designed for rosette plants with round or elliptical leaves; it may not perform well for plants with different leaf shapes. The accuracy depends on the quality of input images and may be affected by severe occlusions or non-green elements. Computational time is higher for high-resolution images (e.g., A3 dataset).
1：Experimental Design and Method Selection:
The methodology includes a three-step process: statistical-based image enhancement, graph-based leaf region extraction, and Circular Hough Transform for leaf counting. It uses Weibull distribution for image enhancement, graph algorithms for segmentation, and CHT for counting.
2：Sample Selection and Data Sources:
Benchmark datasets A1, A2, and A3 from the Leaf Segmentation Challenge (LSC) are used, consisting of Arabidopsis and tobacco plant images with varying complexities, resolutions, and backgrounds.
3：List of Experimental Equipment and Materials:
A system with Intel i3 processor 2.66 GHz, 4 GB memory, 64-bit Windows OS, and Matlab (release 2016) software. Images captured using a 7 MP Canon Power-Shot SD1000 camera for A1 and A2 datasets, and Grashopper cameras for A3 dataset.
4：66 GHz, 4 GB memory, 64-bit Windows OS, and Matlab (release 2016) software. Images captured using a 7 MP Canon Power-Shot SD1000 camera for A1 and A2 datasets, and Grashopper cameras for A3 dataset.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Convert RGB images to HSV, enhance the V plane using statistical methods, construct a graph for segmentation, apply graph-based algorithms to identify leaf regions, remove non-leaf regions using color space thresholds, and use CHT for counting leaves.
5：Data Analysis Methods:
Performance evaluated using metrics such as FBD%, DiC, |DiC|, Dice%, Precision, Recall, and Jaccard. Implemented in Matlab with statistical analysis of results.
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
