【摘要】 针对复杂田间环境下杂草形态相似对深度学习模型识别效果的影响,本文以玉米及其主要伴生杂草作为研究对象,提出一种基于双线性卷积神经网络的细粒度杂草识别方法,用于提升作物与杂草识别的准确率。首先,研究对比了常见通用图像分类模型在杂草识别上的表现,选用识别效果较好的VGGNet-19和ResNet-50作为双线性网络的主干结构,以获取更有效的杂草特征,并采用迁移学习的方式训练网络。实验结果表明,该方法在数据集上的识别准确率高达98.5%,高于单一网络模型的识别效果且能够准确地区分具有高相似度的田间杂草,为智能田间除草作业提供高精度的信息支持。更多还原

【Abstract】 To address the impact of weed morphological similarity on the recognition effect of deep learning models in complex field environments,this paper proposes a fine-grained weed recognition method based on bilinear convolutional neural networks for improving the accuracy of crop and weed recognition,taking corn and its major associated weeds as the research object. Firstly,the study compares the performance of commonly used general image classification models on weed recognition,selects VGGNet-19 and ResNet-50,which have better recognition effect,as the backbone structure of the bilinear network to obtain more effective weed features,and uses migration learning to train the network. The experimental results show that the recognition accuracy of the method on the dataset is as high as 98.5%,which is higher than the recognition effect of a single network model. And the method can accurately distinguish field weeds with high similarity,which can provide high-precision information support for intelligent field weeding operations.更多还原