本文利用深度學(xué)習(xí)和特征提取技術(shù)解決植物病害檢測問題。所有測試和實(shí)驗(yàn)都是使用開源數(shù)據(jù)集Plant Village進(jìn)行的。本文的主要工作是實(shí)現(xiàn)三種不同的深度學(xué)習(xí)模型,即Resnet 50,Google Net和VGG16,并找出其中最適合解決分類問題的網(wǎng)絡(luò)模型。眾所周知,世界人口約為70億,農(nóng)作物疾病是世界糧食供應(yīng)的關(guān)鍵問題,而超過90%的人無法使用能夠識別和解決植物病問題的工具或功能。如今,我們生活在一個(gè)由大規(guī)模技術(shù),主要網(wǎng)絡(luò)覆蓋,高端智能手機(jī)以及人工智能的發(fā)現(xiàn)和改進(jìn)所主導(dǎo)的世界中。將高端智能手機(jī)和基于深度學(xué)習(xí)的計(jì)算機(jī)視覺相結(jié)合成為了可能。人們將其定義為“智能手機(jī)輔助疾病診斷”。在深度學(xué)習(xí)領(lǐng)域,學(xué)者們已經(jīng)訓(xùn)練了多種架構(gòu)模型,其中一些模型的性能達(dá)到了99.53%以上。先前的研究是對每種模型分別進(jìn)行的,每個(gè)模型都產(chǎn)生自己的結(jié)果。但是,在我們的研究中,我們使用最新的技術(shù)將三個(gè)先前測試過的深度學(xué)習(xí)模型（Resnet50,Google Net,VGG16）和兩個(gè)分類器（SVM和KNN）組合在一起,以便比較獲得的結(jié)果并找出哪種模型能夠更準(zhǔn)確且更好地解決植物病害分類問題。在本文中,我們解決了這個(gè)問題。... 

【文章來源】：大連理工大學(xué)遼寧省 211工程院校 985工程院校 教育部直屬院校

【文章頁數(shù)】：59 頁

【學(xué)位級別】：碩士

【文章目錄】：
摘要
Abstract
1 Introduction
    1.1 Overview
    1.2 Definition of Plant Diseases Detection
    1.3 Why is it a real world problem?
    1.4 Objectives and Contributions
    1.5 Thesis Structure
2 Related Work
    2.1 Previous Works on Plant diseases detection
3 Background
    3.1 General Neural Networks
        3.1.1 Convolution
        3.1.2 Max Pooling
    3.2 Recurrent Neural Networks
        3.2.1 Image processing in smart agriculture
        3.2.2 Pre-processing
        3.2.3 Segmentation
        3.2.4 Crop detection
        3.2.5 Use of tracking algorithms
        3.2.6 Plant disease classification using deep learning
    3.3 Overfitting
4 Proposed Method
    4.1 Introduction
    4.2 System Architecture
        4.2.1 Deep feature extraction architecture
        4.2.2 Transfer Learning architecture
    4.3 Data collection and dataset preparation
        4.3.1 Data Collection
        4.3.2 Dataset preparation
    4.4 Data preprocessing
    4.5 Pre-trained CNN models and deep learning networks
        4.5.1 VGG16 network
        4.5.2 Google net network
        4.5.3 Resnet50 network
    4.6 Classification algorithm
        4.6.1 Support vector machine( SVM)
        4.6.2 K-Nearest neighbor(KNN)
    4.7 Performance and evaluation metrics
    4.8 Equipment’s configuration and libraries
5 Result and discussion
    5.1 Feature extraction results with Resnet50,Google Net and VGG
    5.2 Deep learning results based on Resnet50,Google Net and VGG16
    5.3 Results based on traditional shallow Networks
    5.4 Discussion
6 Conclusion and Future Work
References
Research Projects and Publications in Master Study
Acknowledgements



本文編號：3467314