基于随机森林算法的黄瓜种子腔图像分割方法

针对黄瓜表型测量中图像识别问题,为解决黄瓜种子腔与果肉图像灰度差别不大情况下的分割难题,提出了基于随机森林算法(Random Forest,RF)的黄瓜种子腔图像分割方法。首先,通过颜色空间变换,提取样本在RGB、HSV、YCb Cr模型下的9个颜色分量;接着,基于灰度共生矩阵提取样本的能量、熵、对比度、相关性的均值与标准差等8个纹理特征。结合纹理与颜色特征,运用随机森林算法构建像素分类器,实现了种子腔的粗分割。为了提高分割质量,对粗分割的图像进行形态学处理得到最终分割图像。最后,与K-均值聚类(Kmeans)算法、支持向量机(Support Vector Machine,SVM)算法做对比。实验表明:随机森林分割算法正确识别率高达95%,错误识别率在10%之内,处理时间1.6 s左右,分割质量上优于其它两种算法。

Image Segmentation of Cucumber Seed Cavity Based on the Random Forest Algorithm

For identifying regions of interest in the measurement of cucumber phenotype, the segmentation of cucumber seed cavity is difficult because the gray scale difference between cucumber seed cavity and flesh is not obvious. A method based on the Random Forest ( RF) algorithm for image segmentation of cucumber seed cavity was proposed. First, 9 color features were taken from 3 color spaces including RGB, HSV and YCbCr. Then 8 texture features which are com-posed of means and standard deviations of angular second moment, entropy, contrast and correlation based on Gray-level Co-occurrence Matrix were extracted. All color and texture features were brought into the training module of Random Forest to make an image classifier which can be used for coarse segmentation of cucumber seed cavity. Morphological pro-cessing was introduced to improve the segmentation quality. Finally, the method mentioned above was compared with the K-means clustering ( K-means) algorithm and the Support Vector Machine ( SVM) algorithm. The test results show that the rate of right identification is up to 95% or more, the rate of false identification is less than 10% and the recognition time is about 1. 6 s by using the Random Forest algorithm for cucumber seed cavity segmentation, which is superior to the other two algorithms.