实蝇图像识别中的形态特征提取研究

Math-morphological Feature Extraction in Classification of Tephritidae

研究特征提取是图像识别中的重要环节,为更加准确地对云南省常见的有害果蔬实蝇进行分类,针对实蝇昆虫的特点,对形态学特征的提取、测量和选择进行了研究。分别对实蝇整体和躯干部分提取了16种形态特征作为原始特征,包括新提出的描述昆虫斑点面积比例的空心度和描述昆虫体型比例的质心距离比的定义。采用一种改进的基于支持向量机(SVM)的特征选择方法并应用此方法对原始特征进行筛选,根据各维特征的贡献大小最终得到9个最优特征。进行仿真的结果显示,对实蝇识别分类取得了良好的效果,有利于后期昆虫分类。

The feature extraction is an important part in image recognition.The sixteen morphological features were extracted from the images in order to classify Tephritidae in Yunnan,including the definition of two new features.An improved feature selection approach based on support vector machine was used in this paper.The algorithm selected nine features from the original features to composes the optimal feature space.The experimental results show that the feature extraction of Tephritidae is practical and feasible.