摘要
深度学习技术能以端对端方式实现农作物害虫识别,克服了传统机器学习方法特征选择具有主观性以及提取特征操作繁琐等不足,但识别的准确率和鲁棒性仍有待提高。为了研究出一种快速,高效的水稻害虫识别方法,本研究以稻纵卷叶螟、三化螟、稻蝗、稻飞虱4种常见的水稻害虫为研究对象,对传统的卷积神经网络Alexnet进行优化改进。首先从自然环境以及搜索引擎上获取4种不同的水稻害虫图像,并对图像进行数量扩增和细节增强预处理。然后对传统的卷积神经网络Alexnet进行优化改进,在Alexnet模型基础上,去除原有局部响应归一化层,在每一个卷积层后加入批归一化层,并采用全局平均池化和激活函数PReLU对模型结构进行优化。结果表明:改进后的模型在害虫数据集上的识别率不低于98%,相比于原网络提升了1.96%,高于LeNet5、VGG13、VGG16等传统网络;改进后的模型的损失值稳定在0.03附近,相比于原网络降低了0.1,均低于LeNet5、VGG13、VGG16等传统网络。从实验结果来看,改进后的方法在水稻害虫分类上有较高的识别率和较好的鲁棒性,可以为农作物害虫的智能识别提供了新的思路和方法。
Deep learning technology can realize crop pest recognition in an end-to-end way,overcoming the shortcomings of traditional machine learning methods such as subjectivity of feature selection and tedious operation of feature extraction,but the accuracy and robustness of recognition still need to be improved.In order to develop a rapid and efficient identification method for rice pests,the four common rice pests,rice leaf roller,yellow rice borer,rice grasshopper and rice planthopper,were taken as the research objects to optimize and improve the traditional convolutional neural network,Alexnet.Firstly,the images of 4 different rice pests were obtained from the natural environment and Internet,and the images were preprocessed by methods of quantity amplification and detail enhancement.Then,the traditional convolutional neural network AlexNet was optimized and improved.Based on the AlexNet model,the original local response normalization layer is removed,and the batch normalization layer is added after each convolutional layer,and the global average pooling and activation function PReLU are used to optimize the model structure.The experimental results show that:The accuracy of the improved model on the pest data set is not less than 98%,which is 1.96%higher than the original network and higher than the traditional network such as LeNet5,VGG13 and VGG16.The loss value of the improved model is stable around 0.03,which is 0.1 lower than the original network and lower than the traditional network such as LeNet5,VGG13 and VGG16.The experimental results show that the improved method has a higher accuracy and better robustness in the classification of rice pests,which can provide a new idea and method for the intelligent identification of crop pests.
作者
肖小梅
杨红云
易文龙
万颖
黄琼
罗建军
XIAO Xiao-mei;YANG Hong-yun;YI Wen-long;WAN Ying;HUANG Qiong;LUO Jiang-jun(School of Software, Jiangxi Agricultural University, Nanchang 330045, China;School of Computer Science and Information, Jiangxi Agricultural University, Nanchang 330045, China)
出处
《科学技术与工程》
北大核心
2021年第22期9447-9454,共8页
Science Technology and Engineering
基金
国家自然科学基金(61562039,61762048)。
