摘要
针对作物叶部病斑区域图像边界模糊和不确定性等因素,以大豆病叶为对象,提出采用遗传神经网络对叶片病斑进行分割的方法,引入遗传算法优化神经网络的权值和阈值,提高了网络训练速度,避免了传统BP算法的局部最小值。通过对大豆灰斑病病斑图像分割的实验表明,该方法速度快且稳定性好,精度高且鲁棒性好。
To solve the problem of crop leaf disease image with indistinct or uncertainty boundary,this paper presents to use threshold segmentation method to separate soybean disease leaves from the background,and then to use genetic neural network approach to separate the lesion from the leaves.The use of genetic algorithm helps to realize structure design and weight value learning of neural network of mottle segmentation.This method can overcome the disadvantages of human beings＇ defining structure when designing neural networks in the past,and can effectively shorten the time of neural network design and training.Experiments show that by using this method to segment mottle with high efficiency,it can fully meet the requirements of soybean leaf disease diagnosis.
出处
《自动化技术与应用》
2013年第11期11-14,23,共5页
Techniques of Automation and Applications
基金
黑龙江省青年科学基金项目(QC2010077)
哈尔滨市创新人才研究专项资金(2013RFQXJ033)
关键词
图像分割
叶片病斑
大豆
神经网络
image segmentation
leaf disease
soybean
neural network
