基于PLSA和颜色命名的小麦图像分割方法

Segmentation Method of Wheat Image Using PLSA and Color Naming

为减少大田环境下光照不足对小麦图像分割的影响,以及提升小麦图像中偏黄叶片的提取效果,提出了将白平衡调整、局部同态滤波预处理和基于概率潜在语义分析(PLSA)模型的颜色命名算法相结合用于小麦图像分割的方法。首先,对大田采集的小麦图像进行白平衡调整,得到准确无偏色的图像;然后对光照不足的图像在HSI彩色模型下对亮度分量I进行局部同态滤波处理,以减少光照不足对图像的影响;最后在RGB彩色模型下基于PLSA模型构建的颜色名RGB值字典,提取图像中绿色和黄色像素点对应区域作为目标区域。结果表明,经白平衡调整后F1值提高1.61个百分点;光照不足图像经局部同态滤波处理后F1值提高12.43个百分点,分割效果明显提升;所提方法对绿色、叶片偏黄及光照不足的小麦图像分割的F1值分别为96.39%、97.29%和96.22%,均达到了较好的分割效果;所提方法与K-means聚类算法相比,虽点状噪音和细小孔洞相对较多,但在分割叶片偏黄小麦上F1值提高4.42%,整体分割效果较好,且稳定性强。

In order to reduce the influence of insufficient light and improve the extraction effect of slightly yellow leaves in the segmentation of wheat image in the field environment,a new method of wheat image segmentation combining white balance adjustment,local homomorphic filtering preprocessing and color naming algorithm based on probabilistic latent semantic analysis(PLSA)model was proposed.Firstly,the wheat image collected in the field was adjusted by the white balance to get the image without color deviation;then the local homomorphic filtering of luminance component I in HSI color model was executed for reducing the influence of insufficient light on the image;finally,the corresponding areas of green and yellow pixels from RGB image were extracted as the target area using the color name RGB value dictionary constructed by PLSA model.The results showed that the F1 value increased by 1.61 percentage points after the white balance adjustment;the F1 value of the insufficient light image segmentation increased by 12.43 percentage point after the local homomorphic filtering,and the effect of segmentation was improved obviously;the F1 values of green,slightly yellow and insufficient light wheat images were respectively 96.39%,97.29%and 96.22%by the method,which achieved good segmentation effect;compared with K-means clustering algorithm,although there were more point noise and small holes in this method,the F1 value for slightly yellow wheat image was increased by 4.42%,the overall segmentation effect was better and the stability was stronger.