 |
|
|
|
以作者查詢圖書館館藏 、以作者查詢臺灣博碩士 、以作者查詢全國書目 、勘誤回報 、線上人數：4 、訪客IP：3.145.58.169
姓名 吳永璿(Yung-Syuan Wu) 查詢紙本館藏 畢業系所 資訊工程學系 論文名稱 應用分類重建學習偵測航照圖幅中的新穎坵塊
(Detecting Novel Parcels in Aerial Images Using Classification Reconstruction Learning)檔案 [Endnote RIS 格式]
[相關文章]
[文章引用]
[完整記錄]
[館藏目錄]
至系統瀏覽論文 ( 永不開放)
摘要(中) 彩色紅外線(英語：Color-Infrared, 縮寫 CIR)航照圖幅中的水稻坵塊識別可揭示農作物種植園的位置和面積等資訊，從而協助政府制定政策。先前針對水稻坵塊識別的研究[1]已經為此任務實現了一個封閉集合分類器，並且它被證明在推理論過程中容易受到新穎坵塊的影響而產生誤判。該封閉集合分類器由於缺乏分布外檢測，在推論階段是不穩健的。本文提出了一種基於開放集合識別分類重建學習 (英語：Classification Reconstruction Learning for Open Set Recognition，縮寫CROSR) 的新穎坵塊檢測方法。具體來說，我們用名為 RecUNet 的特製深度神經網絡替換了原 CROSR 架構中用於提取特徵向量的神經網絡 DHRNet，RecUNet 有助於保存低階語義密集特徵圖。實驗證明 RecUNet 相較於 DHRNet 可以在 AUROC 效能評估指標帶來百分之十五的提升。此外，我們提供的初步結果顯示在航照圖幅內檢測到新穎坵塊時，使用每個航照圖幅中的新穎坵塊檢測率來預測逐圖幅水稻識別性能下降是可行的。 摘要(英) Rice parcel recognition in Color Infrared (CIR) aerial images reveals information such as location and area of crop plantations that assist government policymaking. Prior work on rice parcel recognition[1] has implemented a closed-set classifier for this task, and it was shown to be vulnerable to novel parcels during inference. Without outlier a detection mechanism, the close set classifier is not robust during inference. This paper presents an approach for novel parcel detection in CIR aerial image frames based on Classification-Reconstruction learning for Open-Set Recognition (CROSR). Specifically, we replaced the neural network for latent vector extraction DHRNet with a customized
deep neural network named RecUNet that facilitates preservation of low-level semantic
dense representation maps. Experimental validation demonstrated that RecUNet brought
an improvement of 15 percent in terms of AUROC compared to DHRNet. Furthermore,
we provide preliminary findings showing that upon detection of novel parcels within aerial
image frames, it is feasible to predict frame-wise rice recognition performance drops using
per aerial image frame novel parcel detection ratios.關鍵字(中) ★ 分布外檢測
★ CROSR
★ 坵塊分類
★ 航攝影像關鍵字(英) ★ OOD detection
★ CROSR
★ parcel recognition
★ aerial images論文目次 摘要 (p.v)
Abstract (p.vi)
Contents (p.vii)
1 Introduction (p.1)
1.1 Background (p.1)
1.1.1 Aerial Images and Parcel Maps (p.1)
1.1.2 Novel Parcels (p.3)
1.1.3 Abnormal Frames (p.3)
1.2 Motivation (p.3)
1.3 Problem Description (p.4)
1.4 Objective (p.5)
1.5 Contribution (p.6)
1.6 Thesis Structure (p.6)
2 Related Work (p.7)
2.1 Rice Parcel Recognition (p.7)
2.1.1 VGG16BN-G (p.7)
2.2 Out of Distribution (OOD) detection (p.7)
2.2.1 Maximum over Softmax Probabilities (MSP) (p.8)
2.2.2 Class Belongingness Modelling (p.8)
2.2.3 Openmax (p.9)
2.2.4 Classification reconstruction learning (CROSR) (p.9)
2.2.5 Open-Set Classification in Remote Sensing Hyperspectral Images (p.10)
3 Proposed Method (p.11)
3.1 Novel Parcel Detector (p.11)
3.1.1 Data preprocessing (Parcel Extraction) (p.11)
3.1.2 RecUNet (p.12)
3.1.3 Openmax (p.13)
3.2 The Frame-wise Performance Drop Detector (p.14)
4 Experiments and Results (p.16)
4.1 Dataset (p.16)
4.1.1 Study Area and Aerial Image Frames (p.17)
4.1.2 Parcel Maps (p.17)
4.1.3 Rice Recognition Labels (p.17)
4.1.4 Train Test Splits (p.18)
4.1.5 Description of the Known Novel Parcels (p.19)
4.1.6 Labeling Abnormal Frames (p.20)
4.2 Evaluation Metrics (p.20)
4.2.1 AUROC (p.20)
4.2.2 Recall and Precision (p.21)
4.2.3 Accuracy (p.21)
4.2.4 Parcel based kappa with area weights (p.21)
4.3 Experiment 1 : Evaluating the Novel Parcel Detector (p.21)
4.3.1 Objective (p.21)
4.3.2 Experiment Design (p.22)
4.3.3 Experimental results (p.23)
4.4 Experiment 2 : Evaluating the Frame-wise Performance Drop Detector (p.23)
4.4.1 Objective (p.23)
4.4.2 Experiment Design (p.24)
4.4.3 Experimental results (p.25)
4.5 Experiment 3 : Ablation Studies for the Novel Parcel Detector (p.27)
4.5.1 Objective (p.27)
4.5.2 Experiment Design (p.27)
4.5.3 Experimental results (p.29)
4.6 Experiment 4 : Generalization Test for the Frame-wise Performance Drop Detector (p.33)
4.6.1 Objective (p.33)
4.6.2 Experiment Design (p.33)
4.6.3 Experimental results (p.34)
4.7 Experiment 5 : Evaluating the Frame-wise Performance Drop Detector using Spatially and Temporally Similar Data (p.35)
4.7.1 Objective (p.35)
4.7.2 Train Test split (p.36)
4.7.3 Experiment Design (p.37)
4.7.4 Experimental results (p.39)
4.8 Experiment 6: Evaluating the Frame-wise Performance Drop Detector using Artificially Generated Abnormal Frames (p.39)
4.8.1 Objective (p.39)
4.8.2 Manually Generated Abnormal Frames (p.39)
4.8.3 Experiment Design (p.40)
4.8.4 Experimental results (p.41)
5 Conclusions and Future Works (p.42)
5.1 Conclusions (p.42)
5.2 Future works (p.42)
References (p.44)參考文獻 [1] C.-W. Chen, “Application of convolutional neural networks to aerial images for parcelbased rice interpretation,” Master’s Thesis, National Central University, 2022.
[2] R. Yoshihashi, W. Shao, R. Kawakami, S. You, M. Iida, and T. Naemura, “Classificationreconstruction learning for open-set recognition,” in IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 4011–4020. doi: 10.1109/CVPR.2019.00414.
[3] S. Liu and W. Deng, “Very deep convolutional neural network based image classification using small training sample size,” in 2015 3rd IAPR Asian Conference on Pattern Recognition (ACPR), 2015, pp. 730–734. doi: 10.1109/ACPR.2015.7486599.
[4] D. Hendrycks and K. Gimpel, “A baseline for detecting misclassified and out-of-distribution examples in neural networks,” in International Conference on Learning Representations,2017.
[5] E. Rudd, L. Jain, W. Scheirer, and T. Boult, “The extreme value machine,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 40, pp. 762–768, Jun. 2015. doi: 10.1109/TPAMI.2017.2707495.
[6] A. Bendale and T. E. Boult, “Towards open set deep networks,” in IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 1563–1572. doi: 10.1109/CVPR.2016.173.
[7] S. Liu, Q. Shi, and L. Zhang, “Few-shot hyperspectral image classification with unknown classes using multitask deep learning,” IEEE Transactions on Geoscience and Remote Sensing, vol. 59, no. 6, pp. 5085–5102, 2021.doi:10.1109/TGRS.2020.3018879.
[8] W. Liu, X. Nie, B. Zhang, and X. Sun, “Incremental learning with open-set recognition for remote sensing image scene classification,” IEEE Transactions on Geoscience and Remote Sensing, vol. 60, pp. 1–16, 2022. doi:10.1109/TGRS.2022.3173995.
[9] D. Pal, S. Bose, B. Banerjee, and Y. Jeppu, “Extreme value meta-learning for few-shot open-set recognition of hyperspectral images,” IEEE Transactions on Geoscience and Remote Sensing, vol. 61, pp. 1–16,2023. doi:10.1109/TGRS.2023.3276952.
[10] Export training data for deep learning, https://pro.arcgis.com/en/pro-app/latest/tool-reference/spatial-analyst/export-training-data-for-deep-learning.htm, Accessed: 2023-12-19.43
[11] O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in Lecture Notes in Computer Science, vol. 9351, Oct. 2015, pp. 234–241, isbn: 978-3-319-24573-7. doi: 10.1007/978-3-319-24574-4_28.
[12] W. J. Scheirer, A. Rocha, R. Michaels, and T. E. Boult, “Meta-recognition: The theory and practice of recognition score analysis,” IEEE Transactions on Pattern Analysis and
Machine Intelligence (PAMI), vol. 33, pp. 1689–1695, 8 2011.指導教授 梁德容 王尉任 張欽圳 林家瑜(Deron Liang Wei-Jen Wang Chin-Chun Chang Chia-Yu Lin) 審核日期 2024-1-19 推文 plurk
funp
live
udn
HD
myshare
netvibes
friend
youpush
delicious
baidu
網路書籤 Google bookmarks
del.icio.us
hemidemi
facebook
twitter
google
reddit