[논문 리뷰] Wheat Crop Yield Prediction Using Deep LSTM Model

Sagarika Sharma, Sujit Rai외 1명|arXiv (Cornell University)|2020. 11. 03.

Remote Sensing in Agriculture참고 문헌 16인용 수 32

한 줄 요약

The paper proposes a CNN-LSTM model that predicts tehsil-level wheat yields in India directly from raw multi-spectral satellite imagery, incorporating contextual land-use data, and achieving substantial improvements over baselines.

ABSTRACT

An in-season early crop yield forecast before harvest can benefit the farmers to improve the production and enable various agencies to devise plans accordingly. We introduce a reliable and inexpensive method to predict crop yields from publicly available satellite imagery. The proposed method works directly on raw satellite imagery without the need to extract any hand-crafted features or perform dimensionality reduction on the images. The approach implicitly models the relevance of the different steps in the growing season and the various bands in the satellite imagery. We evaluate the proposed approach on tehsil (block) level wheat predictions across several states in India and demonstrate that it outperforms existing methods by over 50\%. We also show that incorporating additional contextual information such as the location of farmlands, water bodies, and urban areas helps in improving the yield estimates.

연구 동기 및 목표

• 농민과 기관을 돕기 위한 현장 내 수확량 예측의 필요성 제고와 비용 효율성 확보.
• 수작업 특성 없이 원시 위성 영상만을 사용하는 파이프라인 개발.
• 예측 모델에 맥락 정보(수역, 농지, 도시 지역) 통합.
• 일곱 개 Indian 주의 tehsil 수준 밀 데이터에 대해 방법을 평가하고 기준 방법과 비교.

제안 방법

• 각 tehsil마다 24개월 수의 다중스펙트럴 이미지를 처리하는 3-모듈 CNN-LSTM 아키텍처를 사용한다.
• CNN 특징 추출기: 각 16개의 필터를 가진 5개의 합성곱 층, 3x3 커널, 스트라이드 2, LeakyReLU, 풀링 없음, 출력은 1024 특징으로 펼쳐짐.
• LSTM 시간 인코더: 각 512 노드의 3개 중첩 층과 드롭아웃(75% 유지), 그 뒤에 3층 완전 연결 수율 예측기.
• 각 시점에서 계산된 L2 손실로 학습하여 단계별 출력의 평균으로 조기 시즌 수확량 예측 가능.
• 맥락 대역(수계, 농지, 도시) 포함 여부를 선택적으로 설정하거나 이미지 만으로 학습하는 (CNN-LSTM-9) 구성을 사용해 맥락 영향 평가.

실험 결과

연구 질문

• RQ1 Can a deep CNN-LSTM model trained on raw satellite imagery predict wheat yield at the tehsil level in India without hand-crafted features?
• RQ2 Does incorporating land-use contextual information (water bodies, agricultural land, urban areas) improve prediction accuracy?
• RQ3 How does the CNN-LSTM-12 model compare to traditional NDVI/VCI-based methods and prior deep learning approaches?
• RQ4 Is early-season yield prediction feasible with acceptable accuracy using partial seasonal image sequences?
• RQ5 How well do state-specific models generalize to other states in India?

주요 결과

• CNN-LSTM-12 substantially outperforms baselines using handcrafted features by over 70% and outperforms LSTM+Gaussian Process by over 54%.
• The inclusion of contextual information (water bodies, farmlands, urban areas) improves RMSE by more than 17% on average across states.
• Early predictions improve as more images from the growing season are available, with notable improvements up to step 8 (about two months) before stabilizing.
• State-wise modeling yields better performance than a single multi-state model due to regional heterogeneity.
• Predicted tehsil yields are generally underestimates for large tehsils, with a predominantly linear correlation between predicted and actual yields.