基于卷积神经网络的可见光图像农作物病虫害的检测

发布时间：2018-06-07 16:51

本文选题：农作物病虫害检测 + 卷积神经网络　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：农业生产是我国经济发展至关重要的部分,其产量不仅会影响到国家的经济发展,同时也会影响到人们日常生活的质量。然而,由于传统的农作物病虫害检测过多依赖于专家的经验,效率低且不具有智能性是影响其病虫害检测性能的重要限制。随着计算机视觉以及神经网络算法的不断发展,随着智能手机的不断普及,农作物病虫害的智能检测越来越多地引起了人们的关注,并得到了快速的发展。本文提出了一种基于卷积神经网络的可见光图像农作物病虫害的检测方法,对于农作物病虫害的分类识别有很好的效果。本文提出的基于卷积神经网络的可见光图像农作物病虫害的检测方法,主要结合了:卷积神经网络、迁移学习思想、支持向量机分类以及数据扩充四个方面的技术。本文也针对这四方面技术进行了详细的阐述。在介绍该方法之前,本文对实验部分要用到的农作物病虫害数据集做了详细的介绍,同时介绍了一种传统方法用在该数据集上的分类结果,可以看出传统方法对于数据的分类并不理想。在此之后,本文开始针对检测方法进行介绍。文章首先介绍了仅基于卷积神经网络的检测方法。该部分主要针对卷积神经网络的结构设计、卷积神经网络的训练细节等做了详细的阐述,并列举了仅基于卷积神经网络的检测方法对于农作物病虫害数据集的分类情况,并分析了其产生过拟合问题的原因。随后介绍了基于迁移学习思想对于检测方法的改进。可以看出,迁移学习思想能够较大程度地缓解过拟合问题,使分类检测精度有了很大的提升。在此之后,我们用支持向量机取代原先的softmax分类层进行分类,可以看出这样做可以使准确率小幅度提升,对于缓解过拟合问题有一定的作用。最后,介绍了数据扩充对于缓解过拟合问题、提升病虫害检测精度的重要意义,并给出相关的验结论。此外,本文还将提出的该农作物病虫害的检测方法应用于自己建立的数据集,也可以实现较高的分类精度,说明该方法具有较好的稳健性。本文提出的基于卷积神经网络的可见光图像农作物病虫害的检测方法,相比于传统方法而言,对于病虫害检测具有良好的检测性能,其智能性和高准确性对于实际问题的解决有很好的效果。
[Abstract]:Agricultural production is a vital part of the economic development of our country. Its output will not only affect the economic development of the country, but also affect the quality of people's daily life. However, because the traditional crop pest detection depends too much on the experience of experts, low efficiency and lack of intelligence is an important limitation affecting the performance of crop pest detection. With the development of computer vision and neural network algorithms, and with the popularity of smart phones, the intelligent detection of crop diseases and insect pests has attracted more and more attention, and has been rapidly developed. In this paper, a method of crop pest detection based on convolution neural network is proposed, which has a good effect on the classification and recognition of crop pests and diseases. In this paper, the method of detecting crop pests and diseases in visible light image based on convolution neural network is proposed, which mainly combines four techniques: convolution neural network, migration learning idea, support vector machine classification and data expansion. This article also has carried on the detailed elaboration to these four aspects technology. Before introducing this method, this paper introduces the data set of crop diseases and insect pests used in the experiment in detail, and introduces the classification results of a traditional method used in the data set. It can be seen that the traditional method for data classification is not ideal. After this, this article begins to carry on the introduction to the detection method. Firstly, the detection method based on convolution neural network is introduced. In this part, the structure design of convolutional neural network and the training details of convolutional neural network are described in detail, and the classification of crop diseases and insect pests data set based only on convolution neural network detection method is listed. The reason of over-fitting problem is analyzed. Then the improvement of detection method based on transfer learning is introduced. It can be seen that the idea of transfer learning can alleviate the problem of over-fitting to a great extent and improve the accuracy of classification and detection greatly. After that, we use support vector machine instead of the original softmax classification layer to classify. It can be seen that this method can improve the accuracy slightly, and has a certain role in alleviating the problem of over-fitting. Finally, the importance of data expansion in alleviating the problem of over-fitting and improving the accuracy of disease and pest detection is introduced, and the relevant test results are given. In addition, the method proposed in this paper is applied to the data set established by ourselves, and it can also achieve higher classification accuracy, which shows that the method has good robustness. In this paper, a new method based on convolution neural network is proposed to detect crop diseases and insect pests in visible light images. Compared with traditional methods, it has a good performance in detecting pests and diseases. Its intelligence and high accuracy have a good effect on the solution of practical problems.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S432;TP391.41

【参考文献】