基于改进K-means图像分割算法的细叶作物覆盖度提取

Improving Accuracy of Fine Leaf Crop Coverage by Improved K-means Algorithm

摘要:

植被覆盖度是重要的农学指标，图像法作为一种方便、快捷、准确度较高的地面测量方法，在该领域得到了广泛应用。图像背景分割是获取植被覆盖度最关键的步骤，已有分割算法的分割对象局限于大叶植物或者长势较为稀疏的作物，针对细叶作物的研究较少，或者未根据分割结果得出更有价值的规律。本文以小麦为例，提出了基于HSV空间的自适应果蝇均值聚类算法（IFOA-K-means），用来分割图像背景，以此作为获取覆盖度变化的理论基础。采用小波分析按比例去噪算法单独对亮度分量去噪，主体分割算法采用自适应步长果蝇算法（IFOA）改进的K-means算法对小麦图像进行背景分割，综合了自适应果蝇算法的全局最优和K-means算法的局部最优特点，使分割效果达到最优。其分割效果优于基于遗传算法的最大类间方差分割法，较好地去除了滴灌带等较明显干扰因素，与传统的K-means算法相比，运行时间和峰值信噪比指标都较优，小麦覆盖度准确率在90%以上，与作物系数之间的决定系数为0.9531。

Abstract:

Canopy coverage is an important agronomic indicator. Image method is widely used in this field as a convenient, fast and accurate ground measurement method. Image background segmentation is the most critical step to obtain canopy coverage. Some segmentation algorithms have been limited to largeleaf plants or crops with relatively sparse growth. Few studies were on fine leaf crops, or no more valuable rules based on segmentation results. Therefore, taking wheat as an example, an IFOA-K-means algorithm based on HSV space was proposed. The K-means algorithm split the image background as a theoretical basis for obtaining coverage changes. Then the wavelet denoising algorithm was used to denoise the luminance component separately. The main segmentation algorithm was improved by the adaptive step size fruit fly algorithm (IFOA). The Kmeans algorithm was used to perform background segmentation on wheat images, and the global optimality of the adaptive fruit fly algorithm and local optimal features of the K-means algorithm were integrated to optimize the segmentation effect. The segmentation effect was better than the Ostu method based on genetic algorithm. It was better to remove the obvious interference factors such as drip irrigation belt, compared with the traditional Kmeans algorithm, the segmentation results were superior to the traditional algorithms in terms of running time and peak signal-to-noise ratio. The accuracy of wheat coverage was over 90%, the fit to the crop coefficient was as high as 0.9531, and the estimation of wheat growth status was estimated.