生态背景下基于人工智能深度学习的竹类害虫识别方法研究

Research on Bamboo Pests Identification Method Based on Artificial Intelligence and Deep Learning under the Ecological Context

针对生态背景下的竹类害虫识别,作者研究了一种基于人工智能深度学习的识别方法。构建了具有5663张图片的虫类数据集,其中包含3种竹类害虫和3种其他虫类,利用深度学习模型GoogLeNet特有的Inception模块构成的网中网结构,使其获得更多的图片特征,并开展了4组不同训练集与测试集比例的实验。结果表明:模型的精确度随训练集比重的增大而增大,当训练集和测试集的比例为9∶1时表现最好,F1值达到了95.48%,模型精确度为97.5%,体现了识别模型具有较好的综合性能和较高的实用性。该方法能较好地实现3种竹类害虫在生态背景下的智能识别,是针对竹类生产经营中的虫害防治问题的一种智能化解决方案,为竹产业精细化管理及高效生产经营提供有效的科技支撑。

This paper studied a bamboo pests identification model based on artificial intelligence deep learning under the ecological context. An insect dataset with 5,663 photos was established,consisting of 3 species of bamboo pests and 3 other insect species. Furthermore,a special " network in network" architecture composed of Inception modules in a deep learning model named GoogLeNet was introduced and applied to extract image features. Meanwhile,this model was run on 4 train-test set split experiments. The experimental results showed that the accuracy of this model increased with the increase of the ratio of the training set. When the ratio of the train-test set was set to 9 ∶ 1,this model achieved the best performance,and when the mean-F1-Score reached 95. 48%,the overall classification accuracy reached 97. 5%. These experimental results implied that this model possessed relatively good comprehensive performance and practicability. This model could realize the intelligent recognition of the 3 species of bamboo pests mentioned above under the ecological context and is an intelligent solution to the pest control in bamboo production and management. Additionally,it could provide effective technology backup for intensive and effective management of bamboo industry.