摘要
针对当前葡萄叶部病害大数据集匮乏、数据集品质较低等问题,提出一种新型卷积神经网络识别模型。首先,通过3×3的卷积核与池化层的交替结构,实施了一个初步的特征抽取策略。随后,通过空洞卷积技术,构建了一个多分支的模块,旨在捕获多尺度下的病斑特征。在网络的深层,引入了稠密连接策略,进一步完善了葡萄叶片病害的分层识别模型。这个模型不仅可以准确识别常见的葡萄叶片病害,还可以根据病害的严重程度提供相应的治疗建议。最后通过实验及仿真证明,所提新型卷积神经网络识别模型方便了种植户准确地诊断葡萄叶片病害情况,有效减少了经济损失。
A novel convolutional neural network recognition model was proposed to address the current problems of lack of large datasets and low quality of datasets for grape leaf diseases.First,a preliminary feature extraction strategy was implemented by alternating the structure of a 3×3 convolutional kernel with a pooling layer.Subsequently,a multi-branching module aimed at capturing lesion features at multiple scales was constructed through a null convolution technique.In the deeper layers of the network,a dense connectivity strategy was introduced to further refine the hierarchical model for recognizing grape leaf diseases.This model not only accurately recognizes common grape leaf diseases,but also provides appropriate treatment suggestions according to the severity of the diseases.Finally,it is demonstrated through experiments and simulations that the proposed novel convolutional neural network recognition model facilitates growers to accurately diagnose grape leaf disease conditions and effectively reduce economic losses.
作者
吴方婧
张华民
WU Fang-jing;ZHANG Hua-min(School of Mechanical Engineering,Anhui Science and Technology University,Fengyang,233100,Anhui;School of Information and Network Engineering,Anhui Science and Technology University,Bengbu,233030,Anhui)
出处
《蚌埠学院学报》
2024年第2期64-70,共7页
Journal of Bengbu University
基金
国家自然科学基金项目(62072323)。
关键词
深度学习
病害防治
特征识别
病斑分割
神经网络
deep learning
disease control
feature recognition
spot segmentation
neural network
