自然条件下葡萄叶片的自动分割的算法研究

Research on Automatic Segmentation Algorithm of Grape Leaf under Natural Condition

关键词:  葡萄  颜色特征  马尔科夫随机场  能量函数  图割

Key words:  grape  color characteristics  markov random field  energy function  graph cut

摘要:

叶片是植物重要组成部分，在植物生长状态和病虫害检测的研究中，叶片区域的分割是至关重要的一步。本文提出了一种具有普适性的自动葡萄叶片分割算法。本算法在马尔科夫随机场的基础上，建立能量函数，用混合高斯模型对目标和背景的概率密度进行估计，通过图割算法分割出目标叶片。本文通过大量统计发现对于葡萄叶片，采用G/R以及a*颜色特征来筛选目标和背景种子具有很高的可靠性，可将其作为自动选择种子点的依据。本文将多种反映绿色的特征作为本算法的分割特征，对它们分割精度进行了测试，发现对于不同时间、不同天气的叶片图像，单一G/R和a*具有最好的效果，分割精度分别达到86.74%和92.38%，若用它们组合为双特征，分割效果会进一步提高，分割精度达95.03%。

Abstract:

Leaf is the important organ of the plant. Leaf information extraction is the crucial step in the research of plant growth status and plant diseases and insect pests detection. On the basis of markov random field, the study built energy function and extracted the G/R as well as a* color attributes of grape leaf, then filtered target and seed points of background automatically, meanwhile the Gaussian mixture model was set up to estimate the probability density of objective and background,then segmented out the leaf information which is eventually needed by graph cut method.The tests shows that the methods utilized G/R and a* attributes to filter seed points have high accuracy and they can be the gist of choosing the seed points thtough largely counting the grape leaves.The paper put many characteristics reflectting the green as segementation feature of the algorithm.It found that singlely using G/R or a* had the better result for leaves image in the different time and weather. The segementation accuracy of them can reach 86.74% and 92.38% respectively.If they are combined into double characteristic,the result of the segementation will be more accurate and the segementation accuracy can reach 95.03%.