| 摘要: |
| 【目的】探究非垂直拍摄获取细叶作物覆盖度优化算法以及非垂直拍摄对作物覆盖度计算结果的影响。【方法】在北京市小汤山基地草业研究中心露天环境下，以青绿薹草、结缕草为试验对象，获取45°、60°、90°（与地面夹角）3种拍摄角度下的图像，采用基于自适应权重粒子群改进K-means的图像分割方法，分析不同拍摄角度对青绿薹草覆盖度测量精度的影响，研究非垂直拍摄与垂直角度下覆盖度的关系曲线。首先将草RGB图像转化成HSV颜色空间，在HSV颜色空间利用自适应权重PSO算法向全局像素解的子空间搜寻，通过迭代搜寻到全局最优解，确定最佳初始聚类中心；其次利用K-means算法对不同角度图像像素进行聚类划分，从而得到草冠层区域分割结果，最后采用形态学滤波方法对分割结果进行优化。【结果】垂直拍摄时，传统K-means算法计算的2个品种的覆盖度总体相对误差分别为32.89%和34.37%，而本文算法下2个品种总体相对误差分别为11.23%和15.85%。相比于K-means算法，本文算法环境适应性好，算法精度高。非垂直拍摄条件下，本文算法能够克服多角度拍摄导致图像色彩分布不均匀的问题，有效分割出草冠层区域，90°覆盖度估测值与真实值平均误差为3.27%，60°二者平均误差为4.61%，45°平均误差为5.70%，随着拍摄角度的减小，平均误差略有增大，但均小于6%。非垂直角度下计算的覆盖度与垂直角度覆盖度呈显著地线性关系。【结论】采用本文方法可以提高非垂直拍摄获取作物覆盖度的精度。 |
| 关键词: 图像分割； K-means算法； PSO优化算法； 多角度； 作物覆盖度 |
| DOI：10.13522/j.cnki.ggps.2019119 |
| 分类号: |
| 基金项目: |
|
| Research on Optimization Algorithm of Fine Leaf Crop Coverage Based on Non-vertical Shooting |
|
ZHANG Hui, WU Pinghui, ZHANG Xin, SUN Tiejun, XUE Xuzhang, ZHENG Wengang
|
|
1. School of Electronic Information Engineering, Hebei University of Technology, Tianjin 300401, China;2. Beijing Research Center for Information Technology in Agriculture, Beijing 100097, China;3. Beijing Grass Industry and Environment Research Center, Beijing Academy of Agriculture and Forestry, Beijing 100097, China
|
| Abstract: |
| 【Objective】 The purpose of this paper is to explore the optimization algorithm of fine leaf crop coverage calculated through non-vertical photography and analyze the influence of non-vertical photography on crop coverage calculation results.【Method】This paper took Carex breviculmis and Zoysia japonica Steud as experimental objects in the open environment of Xiaotangshan Base Grass Research Center in Beijing, and obtained images from three shooting angles of 45°, 60° and 90° (angle with ground). To analyze the influence of different shooting angles on the measurement accuracy of Carex breviculmis coverage, and the relationship between non-vertical shooting and coverage under vertical angle was studied, an improved K-means method using adaptive weight particle swarm optimization (APSO) was proposed to conduct image segmentation. Firstly, the RGB image of grass was transformed into HSV color space. In the HSV color space, the APSO algorithm was used to search the subspace of the global pixel solution. The global optimal solution was found by iterative search to determine the best initial cluster center. Secondly, the k-means algorithm was used to cluster the pixels of different objects to obtain the result of segmentation of the canopy layer. Finally, the segmentation result was optimized by the morphological filtering method. 【Result】 When shooting vertically, the overall relative errors of the two varieties calculated by the traditional k-means algorithm were 32.89% and 34.37% respectively, while the overall relative errors of the two varieties calculated by this algorithm were 11.23% and 15.85% respectively. Compared with k-means algorithm, the algorithm in this paper has good environmental adaptability and high accuracy. Under the condition of non-vertical shooting, the algorithm could overcome the problem of uneven color distribution of images caused by multi-angle shooting, and effectively segment the area of the canopy layer. The average error of estimated value and true value of 90° coverage was 3.27%, and the average error of 60° was 4.61%. The average error of 45° was 5.70%. As the shooting angle decreases, the average error increases slightly, but all of them are less than 6%.The coverage calculated at non-vertical angles has a significant linear relationship with that at vertical angles. 【Conclusion】 The improved K-means method using adaptive weight particle swarm optimization (APSO) can improve the accuracy of crop coverage obtained by non-vertical photography. |
| Key words: image segmentation; K-means algorithm; PSO optimization algorithm; multi-angle; crop coverage |
