自然光环境下马铃薯外部缺陷的无损检测方法研究

发布时间：2018-04-20 21:11

本文选题：自然光环境 + 饱和度灰度化　；参考：《东华理工大学》2017年硕士论文

【摘要】：马铃薯作为我国第四大主粮,对其外部缺陷的检测直接影响到我国马铃薯商品化和深加工进程。由于目前国内大多基于特定环境、特定光照下进行检测,这并不能满足实际环境检测要求。因此,本文提出在开放的自然光环境下对马铃薯主要外部缺陷绿皮、发芽、病斑等的无损检测方法进行研究,这对马铃薯进行深加工和商品化具有重大意义。本文首先针对自然光环境下马铃薯图像的分割方法进行研究,提出一种新的方法分割出马铃薯目标区域。主要先通过R-B和G-B颜色分量差双阈值法来消除马铃薯颜色背景,再通过基于饱和度灰度化结合大津法进行阈值分割,并结合形态学滤波处理去除孔洞等噪声,最终实现自然光环境下大部分样本图像的马铃薯区域分割。然后设计了一个卷积神经网络模型进行分类,主要是通过前面对马铃薯样本图像进行目标分割预处理后,将其输入到该卷积神经网络中进行训练学习,最后成功对正常薯、绿皮薯、发芽薯和病斑薯样本图像分类。最后分别针对缺陷区域的检测方法进行研究。针对绿皮区域提出一种基于RGB、HSV和LAB多颜色模型综合特征的检测方法,主要通过分别提取RGB模型的颜色分量差R-G、HSV模型的色调H和LAB模型的色度A三个最优特征参数进行融合,最后成功实现对不同程度绿皮区域进行检测;针对发芽和病斑区域提出一种基于Laplace算子灰度方差的检测方法,主要通过Laplace锐化处理后分别计算马铃薯目标小区域的平均方差和整个马铃薯目标区域的平均方差的阈值进行比较,最终成功定位出发芽和病斑区域。
[Abstract]:Potato is the fourth main grain in China. The detection of its external defects has a direct impact on the commercialization and deep processing of potato in China. At present, most of our country based on specific environment, specific light detection, this can not meet the requirements of the actual environment detection. Therefore, in the open natural light environment, the non-destructive detection methods of the main external defects of potato, such as green peel, germination and disease spot, are studied, which is of great significance for the further processing and commercialization of potato. In this paper, the method of potato image segmentation in natural light environment is studied, and a new method is proposed to segment potato target area. Firstly, R-B and G-B color component difference double threshold method is used to eliminate potato color background, then threshold segmentation based on saturation grayscale combined with Otsu method is carried out, and morphological filtering is used to remove voids and other noises. Finally, the potato region segmentation of most sample images in natural light environment is realized. Then a convolutional neural network model is designed to classify the potato sample image, which is input into the convolutional neural network for training and learning. Image classification of green tuber, germinated potato and diseased potato samples. Finally, the detection methods of defect area are studied. A detection method based on the synthetic features of RGBV HSV and LAB multi-color model is proposed for green skin region. The three optimal feature parameters of hue H and LAB model of RGB model are extracted respectively by extracting the color component difference of R-GG HSV model and the chromaticity A of LAB model for fusion. Finally, the detection of green skin regions with different degrees is successfully realized, and a detection method based on gray variance of Laplace operator is proposed for germinating and diseased areas. The threshold values of the mean variance of potato target small area and the whole potato target area were calculated by Laplace sharpening, and the germinating and diseased areas were successfully located.
【学位授予单位】：东华理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S532;TP391.41

【参考文献】