Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (9): 176-181.DOI: 10.3778/j.issn.1002-8331.2003-0014

Previous Articles     Next Articles

Identification Model of Pests on Yuluxiang Pear Leaves Based on TACNN

ZHAO Zhiyan, YANG Hua, HU Zhiwei, YU Haiping   

  1. College of Information Science and Engineering, Shanxi Agricultural University, Jinzhong, Shanxi 030801, China
  • Online:2021-05-01 Published:2021-04-29



  1. 山西农业大学 信息科学与工程学院,山西 晋中 030801


In order to solve the problems that Yuluxiang pear has many types of leaf pests, pests spreads quickly, and it takes a long time to manually identify pear leaf pests. To realize the image automatic recognition?of the pests on the Yuluxiang pear leaves under the natural environment, TACNN Convolution?Neural?Network(TACNN) is proposed as the recognition model of pests. This article first analyzes the network structure of the Alexnet model, and uniformly process the pest image of Yuluxiang leaf collected in the field. In order to avoid the over-fitting phenomenon caused by excessively large convolution kernel parameters of the fully connected layer, by optimizing the fully connected layer and setting different neuron nodes and experimental parameters, two insect recognition models, Mid-Alexnet and TACNN, are obtained. The experimental results show that the TACNN has higher recognition accuracy compared with Alexnet and Mid-Alexnet models. The categorized average accuracy rate of TACNN is 81.18%. The TACNN can accurately distinguish three pear pests such as scarabs, psylla chinensis and pear gall midge. The model has good performance in recognition and achieves precise recognition of the pests under natural environment.

Key words: pear leaves, pests, recognition, convolutional neural network


为了解决玉露香梨叶虫害种类多、扩散速度快、人工识别梨叶虫害耗时长的问题,提出了能够在自然环境下对玉露香梨叶虫害图像自动识别的Tiny-Alexnet卷积神经网络(Tiny-Alexnet Convolution Neural Network,TACNN)的虫害识别模型。分析了Alexnet模型的网络结构,并将实地采集的玉露香叶片虫害图像进行统一处理,为避免全连接层卷积核参数过大而产生的过拟合现象,通过优化全连接层,设置不同神经元节点和实验参数,得出了Mid-Alexnet、TACNN两种虫害识别模型。实验结果表明:TACNN较Alexnet和Mid-Alexnet模型有较高的识别准确率,该模型能够有效地提取梨叶虫害特征,类别平均准确率为81.18%,实现了对金龟子、梨木虱、梨瘿蚊三种虫害的准确区分。该模型在玉露香梨叶虫害识别方面具有良好的性能,可实现自然环境下玉露香梨叶虫害的精准识别。

关键词: 梨叶, 虫害, 识别, 卷积神经网络