基于图像处理和蚁群优化的形状特征选择与杂草识别

Using shape features of plant leaf to identify the weed from the crop is an important method for weed recognition. In order to improve the accuracy and efficiency, overlapped leaves were separated through morphology operation and threshold segmentation ba

摘要: 利用叶片形状特征区分杂草和作物是杂草识别的一个重要方法。为了提高杂草识别的精度和效率，通过形态学运算和基于距离变换的阈值分割方法分离交叠叶片，从单个叶片中提取包括几何特征和矩特征的17个形状特征，用蚁群优化（ACO）算法和支持向量机（SVM）分类器进行特征选择和分类识别，选取有利于分类的较优特征并实现特征的优化组合。棉田杂草试验结果表明，该方法能实现分类特征的有效缩减，经优化组合得到的最优特征子集用于杂草识别的准确率达95%以上，识别率高，稳定性好，对识别杂草时如何兼顾准确率和实时性具有参考意义。

Abstract: Using shape features of plant leaf to identify the weed from the crop is an important method for weed recognition. In order to improve the accuracy and efficiency, overlapped leaves were separated through morphology operation and threshold segmentation ba