基于改进C-V模型的棉花叶片目标提取方法

Target extraction of cotton disease leaf images based on improved C-V model

关键词:  棉花叶片  复杂背景  改进C-V模型  全局信息  局部信息  特征提取  图像分割

Key words:  cotton leaf  complex background  improved C-V model  global information  local information  feature extraction  image segmentation

摘要:

为解决自然条件下棉花叶片因其轮廓几何边缘长势不均匀所导致的叶片目标提取不精准问题，提出一种基于改进C-V模型的棉花病害叶部目标提取方法。在传统C-V模型的基础上，将长度惩罚项和符号距离函数的约束能量项引入能量模型中，以达到对演化曲线长度变化的约束目的，从而完成对整幅图像目标特征的提取。本研究算法先对待分割的图像设置初始曲线，并利用高斯滤波算子对待分割图像进行平滑滤波处理，然后根据图像全局灰度信息和局部二值匹配信息建立能量方程，根据其离散化形式，对水平集函数进行演化，并从中提取演化曲线，最后根据水平集函数演化过程所满足的终止条件，输出图像分割结果。按照不同天气条件和不同背景采集了1 200幅棉花叶片样本图像，对本研究算法进行测试。试验结果表明：本研究算法对于晴天、阴天和雨天图像中目标（棉花叶片）轮廓提取准确率分别达到82.23%、82.73%和84.60%。分割结果表明，本研究算法能够对3种天气条件（晴天、阴天、雨天）与4种复杂背景（白地膜、黑地膜、秸秆、土壤）特征混合的棉花叶片图像目标特征轮廓实现准确提取。

Abstract:

In order to solve the problem that target extraction of cotton leaf is not accurate due to the geometric edge grows unevenly under natural conditions, a method for extracting leaf targets of cotton diseases based on improved C-V model is proposed. The algorithm is improved on the basis of conventional model called C-V. The length penalty term and constrained energy term of symbolic distance function are introduced into the energy model, so as to achieve the goal that the length variation of evolution curve is constrained, thus, the extraction process of the target feature for the whole image is completed. First, an initial curve for the segmented images is set by this algorithm, and the segmented images are processed smoothly by Gauss filter operator. Then, the energy equation is established according to the global gray information of the images and the local two value matching information, and the evolution curve is extracted from the evolved level set function according to the discrete form of energy equation. Finally, according to the end condition of the evolution of the level set function, the segmentation result is output. A total of 1 200 images are gathered according to different weather conditions and different backgrounds, and the segmentation performance of the proposed algorithm is evaluated. Experimental results showed that, for the images taken in sunny day, cloudy day and after raining, the contour extraction accuracies of the algorithm are 82.23%, 82.73% and 84.60% respectively. The experimental results demonstrate that, for cotton leaf images that 3 weather conditions (sunny, cloudy, rainy) are mixed with 4 complex backgrounds (white mulch film, black mulch film, straw, soil), its target feature contours can be extracted accurately by this algorithm. It can provide support for the extraction and identification of cotton leaf diseases and insect pests, and provide reference for the extraction of leaves from other plants.