摘要
针对已有的图像处理或卷积神经网络等方法在进行葡萄叶部病害分类时,存在易受病害图像病斑区域大小和复杂背景等影响、不适用于小样本数据集以及在处理高像素的彩色图像数据集时收敛困难等问题,提出一种改进的AlexNet算法,并对葡萄叶部黑腐病、埃斯卡病和褐斑病等3种病害图像及健康叶部图像进行分类识别.在传统AlexNet算法的基础上增加池化层层数对特征进行压缩,去除冗余信息,并选用Leaky ReLU激活函数替换ReLU函数,避免神经元出现“死亡现象”.结果表明,改进的AlexNet算法对葡萄叶部病害的分类准确率达99.1%,明显高于传统AlexNet算法,可为葡萄叶部病害的及时治理提供有效的技术支持.
In view of the existing image processing or convolutional neural network methods in classifying grape leaf diseases,which are affected by the lesion area size and complex background of disease images,not applicable to small sample data sets,and difficult to converge when processing high pixel color image data sets,an improved AlexNet algorithm is proposed.Three grape leaf diseases,namely black rot,Esca and brown spot,and healthy leaf images,are classified and recognized.Based on the traditional AlexNet algorithm,the pooling layer number is added to compress the features,remove redundant information,and the Leaky ReLU activation function is selected to replace the ReLU function,so as to avoid the“death phenomenon”of neurons.The results showed that the improved AlexNet algorithm has a classification accuracy of 99.1%for grape leaf diseases,which is significantly higher than that of the traditional AlexNet algorithm,and can provide effective technical support for timely treatment of grape leaf diseases.
作者
何前
郭峰林
王哲豪
李雅琴
HE Qian;GUO Fenglin;WANG Zhehao;LI Yaqin(School of Mathematics and Computer Science,Wuhan Polytechnic University,Wuhan 430048,China)
出处
《扬州大学学报（自然科学版）》
CAS
北大核心
2023年第2期52-58,共7页
Journal of Yangzhou University：Natural Science Edition
基金
国家自然科学基金资助项目(61906140)。
