基于纹理和梯度特征的苹果伤痕与果梗/花萼在线识别

Online Identification of Apple Scarring and Stems/Calyxes Based on Texture and Edge Gradient Features

为了解决苹果果梗/花萼与伤痕在线识别的问题,利用自行设计的机器视觉检测系统在线采集苹果图像,通过自动分割合成算法将3个不同运动状态下的图像进行合成,使得合成后图像可以包含苹果的整个表面。再利用感兴趣区域提取算法提取出苹果合成图像中的果梗/花萼和伤痕部分。通过分析早期伤痕、中期伤痕和后期伤痕的纹理特征和边缘梯度特征,得出纹理特征适用于早中期伤痕与果梗/花萼的检测,而由于后期伤痕的褐变严重且多已出现凹陷,其纹理特征与果梗/花萼相似,故通过提取后期伤痕和果梗/花萼的边缘梯度特征值用于两者的区分。从SVM的建模结果来看,对于早中期伤痕,模型的总体判别正确率为97%,而后期伤痕的总体判别正确率为96%,并利用所得到的模型设计了用于果梗/花萼与伤痕区分的总体算法。最终通过80个带有不同种类伤痕的样本验证总体算法的正确率为95%,验证试验结果表明该算法可实现对果梗/花萼与伤痕的在线识别。

In order to solve problems relating to online recognition of stems/calyxes and bruise of apples,a self-designed machine vision inspection system was applied to online image acquisition of apples,images of three different motion states were synthesized by the automatic segmentation synthesis algorithm,and stems/calyxes and bruise in images of apples were extracted by the area-of-interest extraction algorithm.To study applicability of different characteristics of images,early bruise mid-term bruise and later bruise were identified through variables of textural features and edge gradient features respectively.As textures of stems and calyxes were more complex than those of early and middle bruise,the support vector machine model based on two variables of textural features,namely entropy and energy/angular second moment,was used and showed a good effect with an overall accuracy of 97%.Due to brown stain and depression of the most later bruises,its textural characteristics were similar to those of stems and calyxes.Hence,later bruise can not be distinguished from stems and calyxes with parameters of textural characteristics.As a result,an edge gradient features extraction algorithm was designed to extract peak intensity and peak positions of later bruises,stems and calyxes and a support vector machine model was created with an overall accuracy of 96%.On this basis,a comprehensive inspection algorithm about stems/calyxes and bruise of apples was designed.Totally 80 different types of bruise-related algorithms were purchased to verify this algorithm and its accuracy reached 95%.Testing results showed that online recognition of stems/calyxes and bruise of apples could be realized through this algorithm.