中文摘要：

采用机器视觉对储粮进行检测是储粮害虫监测的主要方法之一。储粮害虫图像有效分割是粮虫特征提取和识别分类的基础和依据。为实现粮虫图像的有效分割,采用基于图割理论的图像分割方法。该方法采用高斯混合模型（GMM）表征颜色概率分布,在高斯混合模型参数学习估计过程中通过不断扩大背景样本点修正GMM参数,完成对能量函数的最小化,从而改善了分割效果。实验结果表明,该方法可以清晰地分割出与背景差异较小的粮虫,解决了基于直方图模型的图像分割方法无法从目标和背景相似的图像中将粮虫进行准确提取的问题。

英文摘要：

Using machine vision to detect the stored grain is one of the main methods for monitoring stored grain insects. Stored grain insect image segmentation always laid the basis for insects feature extraction and recognition and classification. In order to segment the images effectively, a segmentation method based on graph cuts theory is adopted. In this method, the Gaussian Mixture Model （GMM） is adopted to characterize the color probability distribution. And in the process of GMM parameters learning and estimating, it improved the segmentation results by ontinuous expansion of the sample points, amendment GMM parameters of the background to complete the minimization of energy function. This method is capable of extracting the insects clearly from the similar background ,which is confirmed by the experimental results. It solved the problem of extract the insect objects accurately from the images with objects similar to its background while the gray histogram model based segmentation methods cannot .