RISS 학술연구정보서비스

검색

인기 검색어

    다국어 입력

    http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.

    변환된 중국어를 복사하여 사용하시면 됩니다.

    예시)
    • 中文 을 입력하시려면 zhongwen을 입력하시고 space를누르시면됩니다.
    • 北京 을 입력하시려면 beijing을 입력하시고 space를 누르시면 됩니다.
    닫기

    Language-Enhanced Adaptive Fine-Tuning for Imbalanced Scene Classification in Remote Sensing = 원격탐사 장면 분류에서 불균형 데이터 문제를 위한 언어 강화 적응형 파인튜닝 기법

    한글로보기

    https://www.riss.kr/link?id=T17552474

    • 0

      상세조회
    • 0

      다운로드
    서지정보 열기
    • 내보내기
    • 내책장담기
    • 공유하기
      • URL 복사
    • 오류접수
    인용문이 복사되었습니다.

    부가정보

    목차 (Table of Contents)

    • I Introduction 1
    • 1.1 Remote Sensing Image Classification: Scope and Significance 1
    • 1.2 Fundamental Challenges in Remote Sensing Classification 4
    • 1.2.1 The Domain Gap Between Foundation Models and Remote Sensing 5
    • 1.2.2 Long-Tailed Class Distributions in Remote Sensing 7
    • I Introduction 1
    • 1.1 Remote Sensing Image Classification: Scope and Significance 1
    • 1.2 Fundamental Challenges in Remote Sensing Classification 4
    • 1.2.1 The Domain Gap Between Foundation Models and Remote Sensing 5
    • 1.2.2 Long-Tailed Class Distributions in Remote Sensing 7
    • 1.3 Towards Efficient Domain Adaptation of Foundation Models 11
    • 1.3.1 Parameter-Efficient Fine-Tuning 13
    • 1.3.2 Domain Knowledge Injection 14
    • 1.4 Research Motivation and Problem Statement 16
    • 1.5 Thesis Contributions 18
    • 1.5.1 Bridging the Domain Gap in Forestry Remote Sensing through Lightweight Language-Enhanced Modeling 19
    • 1.5.2 Overcoming Long-Tailed Class Imbalance in Remote Sensing via Ontology-Guided Augmentation and Adaptive Expert Routing 21
    • 1.6 Thesis Structure and Organization 23
    • II Background and Related Work 26
    • 2.1 Remote Sensing Scene Classification and Land Cover/Land Use . 26
    • 2.1.1 Classical Machine Learning Methods 27
    • 2.1.2 Deep Learning for Remote Sensing Scene Classification . 29
    • 2.1.3 Limitations and Contributions 31
    • 2.2 Vision-Language Models for Remote Sensing 33
    • 2.2.1 Large Language Models and Multimodal Learning 33
    • 2.2.2 Vision-Language Models in Remote Sensing 35
    • 2.2.3 Parameter-Efficient Fine-Tuning 37
    • 2.2.4 Instruction-Tuned Datasets for Specialized Domains 39
    • 2.2.5 Limitations and Contributions 41
    • 2.3 Long-Tailed Learning and Class Imbalance in Remote Sensing 42
    • 2.3.1 The Long-Tailed Distribution Problem 42
    • 2.3.2 Data-Level Approaches 44
    • 2.3.3 Loss-Level Approaches 45
    • 2.3.4 Model-Level Approaches: Multi-Expert Frameworks 47
    • 2.3.5 Semantic Knowledge for Long-Tailed Learning 49
    • 2.3.6 Limitations and Contributions 50
    • III Bridging the Domain Gap in Forestry Remote Sensing through Lightweight Language-Enhanced Modeling 53
    • 3.1 Introduction 53
    • 3.2 Methodology 59
    • 3.2.1 Problem Formulation 59
    • 3.2.2 Instruction-tuned Remote Sensing Dataset 60
    • 3.2.3 Small-Scale Vision-Language Model for Deforestation Monitoring 65
    • 3.3 Experimental Results 71
    • 3.3.1 Experimental Settings 73
    • 3.3.2 Accuracy Metrics 74
    • 3.3.3 Multilabel Classification 76
    • 3.3.4 Visual Question Answering 79
    • 3.3.5 Performance Analysis on Rare and Underrepresented Classes 81
    • 3.3.6 Analysis of Deforestation Driven Classes 83
    • 3.3.7 Ablation Study 84
    • 3.4 Discussion and Analysis 86
    • 3.4.1 Training Stability and Convergence Analysis 86
    • 3.4.2 Deployment Efficiency Analysis 87
    • 3.4.3 Broader Implications and Limitations 89
    • 3.5 Conclusion 90
    • IV Overcoming Long-Tailed Class Imbalance in Remote Sensing via Ontology-Guided Augmentation and Adaptive Expert Routing 91
    • 4.1 Introduction 91
    • 4.1.1 Semantic-Aware Augmentation for Long-Tailed Learning 94
    • 4.1.2 Multi-Expert Long-Tailed Recognition 95
    • 4.2 Methodology 97
    • 4.2.1 Problem Formulation 97
    • 4.2.2 Overall Framework 98
    • 4.2.3 Ontology-Guided Semantic Mixing (OGSM) 100
    • 4.2.4 BAGL Module: Adaptive LoRA Expert Ensemble 103
    • 4.3 Experimental Results 106
    • 4.3.1 Experimental Settings 107
    • 4.3.2 Quantitative Results across Benchmarks 109
    • 4.3.3 Tail Class Performance Analysis 110
    • 4.3.4 Generalization on RS Semantic Segmentation 111
    • 4.3.5 Per-Class Performance Comparison 113
    • 4.3.6 Visual Explanation Analysis: Grad-CAM Visualizations . 118
    • 4.3.7 Ablation studies 121
    • 4.4 Conclusion 124
    • V Conclusion 126
    • 5.1 Summary of Contributions 126
    • 5.2 Key Insights and Lessons Learned 128
    • 5.3 Limitations and Open Challenges 130
    • 5.4 Future Research Directions 131
    • 5.5 Broader Impact and Concluding Remarks 133
    • References 135
    • Abstract (In English) 153
    • Abstract (In Korean) 156
    더보기

    분석정보

    View

    상세정보조회

    0

    Usage

    원문다운로드

    0

    대출신청

    0

    복사신청

    0

    EDDS신청

    0

    동일 주제 내 활용도 TOP

    더보기

    주제

    연도별 연구동향

    연도별 활용동향

    연관논문

    연구자 네트워크맵

    공동연구자 (7)

    유사연구자 (20) 활용도상위20명

    해외이동버튼