基于形态学图像处理的麦穗形态特征无损测量

小麦穗部形态参数是直接反应小麦生长状况的重要参数，是育种和考种专家关心的重要参数。为了实现小麦穗部形态特征的无损测量和基于这些特征的快速品种分类，该文提出了基于形态学的穗部性状：芒个数、平均芒长、穗长和穗型的自动提取方法。首先通过小麦图像的形态学运算将麦芒去除得到只有小麦主部的图像，通过寻找主轴方向角和旋转计算外接矩形长度的方法计算穗长，通过对麦芒图像的细化和角点检测方法计算芒长和芒个数，通过宽度系数比例判断穗型，然后利用提取的其中8个特征参数，设计了一个3层的BP神经网络，对4个小麦品种240张图片进行分类识别，识别准确率达到88%。该方法可为小麦快速品种分类提供参考。若能将小麦的其他外部参数同时作为品种识别的输入数据，将会大大提高品种识别的准确性。

Non-destructive measurement of wheat spike characteristics based on morphological image processing

The shape parameter of wheat spike is a direct reflection of the wheat growth status. And it is also an important parameter which the species breeding and test experts care about. In order to achieve the non-destructive measurement of wheat spike morphological characteristics and rapid species classification based on these characteristics, the article proposed the spike traits extraction methods based on morphology: the awn number, the average awn length, the spike length. First, the wheat awn was removed through the wheat image morphological operations so as to get the main image of the wheat. And then calculated the spike length through the method of looking for spindle direction angle and rotating calculation external rectangle length, calculated the awn length and the number of the awn through the method of thinning the wheat awn image and corner detection, and estimated the spike type through width coefficient proportion. Secondly, a three layer BP neural network was designed with eight of the extracted characteristic parameters so as to classify the 240 pictures of 4 wheat varieties. The recognition accuracy rate was 88%. The method can be a reference for rapid species classification of wheat. Taking other external shape characteristics of wheat as supplement input parameters will improve the recognition accuracy greatly.