臺灣博碩士論文加值系統

現代台灣農業面臨人口減少和老化的問題，想到在這樣的環境下，農夫們需要在作物生長期間下田勘查作物的健康狀況肯定相當辛苦。農產業面臨高齡化，在務農、勘災的頻率和效率上也會降低，對於病蟲害的發生若不能提早發現並採取應對措施，則會降低作物收成時的品質與數量，因此我們希望能設計一套能協助農夫了解農作物生長情形，減少人力消耗的系統。本研究透過無人機進行大規模採集農田影像，並對影像資料進行處理，建立了一個自製的數據集，該數據集中包含了1100張圖片，分為3個類別，其類別為正常的葉子、蟲咬的葉子、生病的葉子，其中訓練集包含1000張圖片，驗證集包含100張圖片，我們使用了多種模型進行訓練，如Mask R-cnn、Yolov4、Faster R-cnn等多種模型，並對結果進行比對，最好的結果為Mask R-cnn其準確率(mAP)高達81%，可有效的幫助農民快速知道大範圍農作物目前生長的情況如何，藉此降低作物生長狀況巡檢時所需消耗的人力成本、時間成本。

Considering the declining and aging population of modern Taiwanese agriculture, farmers need to check the health of their crops while they are growing.The agricultural industry is facing aging, and the frequency and efficiency of farming and disaster exploration will also decrease. If the occurrence of diseases and insect pests cannot be detected early and take countermeasures, the quality and quantity of crops will be reduced. Therefore, we hope to design a A system that can help farmers understand the growth of crops and reduce labor consumption.This study collects farmland images on a large scale through drones, processes the image data, and establishes a self-made data set. The data set contains 1100 pictures, which are divided into 3 categories. The categories are normal leaves, insects Bitten leaves, sick leaves, the training set contains 1000 pictures, and the validation set contains 100 pictures, we used a variety of models for training, such as Mask R-cnn, Yolov4, Faster R-cnn and other models, and the results For comparison, the best result is Mask R-cnn, whose accuracy rate (AP) is as high as 81%, which can effectively help farmers quickly know how the current growth of a large range of crops is, thereby reducing the need for inspection of crop growth conditions. Labor cost and time cost.

中文摘要.....i
英文摘要.....ii
誌謝.....iii
目錄.....iv
表目錄.....vii
圖目錄.....ix
第一章 緒論.....1
1.1 研究背景與動機.....1
1.2 研究目的.....2
1.3 研究架構.....2
第二章 文獻探討.....3
2.1 植物蟲害辨識相關論文.....3
2.1.1 害蟲識別數據增強策略.....3
2.1.2 PestNet.....7
2.2 植物病害辨識相關論文.....10
2.2.1 農作物病蟲害檢測.....10
2.2.2 植物病害檢測.....13
2.3 TWO-STAGE DETECTORS.....15
2.3.1 FASTER R-CNN.....15
2.3.2 MASK R-CNN.....18
2.4 ONE-STAGE DETECTORS.....20
2.4.1 YOLOV4.....20
2.4.2 RETINANET.....22
2.4.3 SSD(Single Shot MultiBox Detector).....24
2.5 Latest Detectors.....26
2.5.1 DCNv2(Deformable ConvNets v2).....26
2.5.2 NAS-FPN.....28
第三章 實驗環境.....30
3.1 Detectron2.....30
3.1.1 安裝Detectron2.....30
3.1.2 訓練準備.....31
3.1.3 訓練教學.....34
3.1.4 測試教學.....34
3.2 MMDetection.....35
3.2.1 安裝教學.....35
3.2.2 訓練準備.....35
3.2.3 訓練教學.....36
3.2.4 測試教學.....37
3.3 Darknet.....38
3.3.1 安裝教學.....38
3.3.2 訓練準備.....39
3.3.3 訓練.....40
3.3.4 測試.....40
3.4 CenterNet2.....41
3.4.1 安裝教學.....41
3.4.2 訓練準備.....41
3.4.3 訓練.....42
3.4.4 測試.....42
第四章 實驗方法.....43
4.1 系統架構.....43
4.2 影像資料蒐集.....43
4.3 資料預處理.....44
4.4 深度學習模型訓練.....45
4.4.1 透過資料增強增加樣本數.....45
4.4.2 正常訓練各種模型.....46
4.4.3 修改訓練參數訓練同種模型.....51
4.4.4 修改訓練模型網路層數.....51
4.5 辨識結果.....54
4.5.1 mask辨識結果.....54
4.5.2 bounding box 辨識結果.....56
4.5.3 拼接大圖.....58
4.6 resnet特徵結果圖.....61
第五章 實驗結果.....74
5.1 透過資料增強增加樣本數.....74
5.2 正常訓練模型之結果比較.....74
5.3 修改參數訓練同種模型.....78
5.4 修改模型網路層數.....78
5.5 透過資料增強增加樣本數.....79
第六章 結論.....80
參考文獻.....81

1.行政院農業委員會，資料來源網址: https://www.coa.gov.tw/ws.php?id=2447516
2.行政院農業委員會，資料來源網址: https://www.coa.gov.tw/ws.php?id=2504081
3.Rui Li; Rujing Wang, Jie Zhang, Chengjun Xie, Liu Liu, Fangyuan Wang, Hongbo Chen, Tianjiao Chen, Haiying Hu, Xiufang Jia, Min Hu; Man Zhou, Dengshan Li, Wancai Liu，An Effective Data Augmentation Strategy for CNN-Based Pest Localization and Recognition in the Field， IEEE Access，vol. 7，pp.160274-160283，2019.
4.Liu Liu; Rujing Wang, Chengjun Xie, Po Yang, Fangyuan Wang, Sud Sudirman, Wancai Liu，PestNet: An End-to-End Deep Learning Approach for Large-Scale Multi-Class Pest Detection and Classification， IEEE Access，vol. 7，pp.45301 – 45312，2019.
5.Yong Ai, Chong Sun, Jun Tie, Xiantao Cai，Research on Recognition Model of Crop Diseases and Insect Pests Based on Deep Learning in Harsh Environments， IEEE Access，vol. 8，pp. 171686 – 171693，2020.
6.Shweta Bondre, Ashish K. Sharma，Review on Leaf diseases detection using Deep learning，International Conference on Electronics and Sustainable Communication Systems (ICESC)，September 23，2021.
7.Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun，Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition， IEEE Transactions on Pattern Analysis and Machine Intelligence，vol. 37， no. 9，pp.1904-1916，2015.
8.Ross Girshick，Fast R-CNN，IEEE International Conference on Computer Vision (ICCV)，pp. 1440-1448，September 27，2015.
9.Shaoqing Ren, Kaiming He, Ross Girshick, Jian Sun，Faster R-CNN : Towards Real-Time Object Detection with Region Proposal Networks，IEEE Transactions on Pattern Analysis and Machine Intelligence，vol. 39，no. 6，pp. 1137-1149，2017.
10.Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross Girshick，Mask R-CNN，IEEE International Conference on Computer Vision (ICCV)，pp. 2980-2988，October 29，2017.
11.Tsung-Yi Lin, Piotr Dollár, Ross Girshick, Kaiming He, Bharath Hariharan, Serge Belongie，Feature Pyramid Networks for Object Detection，IEEE Conference on Computer Vision and Pattern Recognition (CVPR)，pp. 936-944，April 19，2017.
12.Alexey Bochkovskiy, Chien-Yao Wang, Hong-Yuan Mark Liao，Scaled-YOLOv4: Scaling Cross Stage Partial Network，IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，pp. 13024-13033，June 25，2021.
13.Zhuliang Yao, Yue Cao, Shuxin Zheng, Gao Huang, and Stephen Lin，Cross-Iteration Batch Normalization，IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) pp. 12326-12335，June 25，2021.
14.Shu Liu, Lu Qi, Haifang Qin, Jianping Shi, and Jiaya Jia，Path Aggregation Network for Instance Segmentation，IEEE/CVF Conference on Computer Vision and Pattern Recognition，June 23，2018.
15.Sanghyun Woo, Jongchan Park, Joon-Young Lee, and In So Kweon，CBAM: Convolutional block attention module，Proceedings of the European Conference on Computer Vision (ECCV)，pp. 3-19，2018.
16.Chien-Yao Wang, Hong-Yuan Mark Liao, Yueh-Hua Wu,Ping-Yang Chen, Jun-Wei Hsieh, and I-Hau Yeh，CSPNet:A new backbone that can enhance learning capability of CNN， IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)，pp. 1571-1580，June 19，2020.
17.Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun，2015，Spatial pyramid pooling in deep convolutional networks for visual recognition， IEEE Transactions on Pattern Analysis and Machine Intelligence，vol. 37, no. 9，pp. 1904-1916，2015.
18.Joseph Redmon and Ali Farhadi，YOLOv3: An incremental improvement，arXiv，April 8，2018.
19.Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár，Focal Loss for Dense Object Detection，IEEE Transactions on Pattern Analysis and Machine Intelligence ，vol. 42，no. 2，pp. 318-327，2020.
20.Mark Sandler, Andrew Howard, Menglong Zhu, Andrey Zhmoginov, Liang-Chieh Chen，MobileNetV2: Inverted Residuals and Linear Bottlenecks，IEEE/CVF Conference on Computer Vision and Pattern Recognition，pp. 4510-4520，June 23，2018.
21.Xizhou Zhu, Han Hu, Stephen Lin, Jifeng Dai，Deformable ConvNets v2: More Deformable, Better Results， IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，pp. 9300-9308，June 20，2019.
22.Golnaz Ghiasi, Tsung-Yi Lin, Ruoming Pang, Quoc V. Le，NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection，IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)，pp. 7029-7038，June 20，2019.
23.Detectron2，https://github.com/facebookresearch/detectron2
24.Detectron2_v0.5，https://github.com/facebookresearch/detectron2/releases
25.Detectron2 Model_Zoo and Baselines，https://github.com/facebookresearch/detectron2/blob/main/MODEL_ZOO.md
26.Detectron2save，https://github.com/kevin0413860616/detectron2save
27.Mmdetection，https://github.com/open-mmlab/mmdetection
28.Darknet-Yolov4，https://github.com/alexeyab/darknet
29.CenterNet2，https://github.com/xingyizhou/CenterNet2
30.CenterNet-Dcnv2，https://github.com/zzzxxxttt/pytorch_simple_CenterNet_45
31.Navaneeth Bodla, Bharat Singh, Rama Chellappa, Larry S. Davis，Soft-NMS -- Improving Object Detection With One Line of Code，IEEE International Conference on Computer Vision (ICCV)，pp. 5562-5570，October 29，2017.