[Paper Review] Label-Efficient Learning in Agriculture: A Comprehensive Review

Jiajia Li, Dong Chen+5 more|arXiv (Cornell University)|May 24, 2023

Smart Agriculture and AI4 citations

TL;DR

This paper presents a comprehensive review of label-efficient learning (LEL) methods in agriculture, categorizing them into weakly supervised (e.g., active, semi-, and weakly supervised learning) and self-supervised/unsupervised approaches. It systematically evaluates their application in precision agriculture, plant phenotyping, and postharvest quality assessment, highlighting key challenges and future research directions to reduce reliance on costly labeled data in agricultural machine learning.

ABSTRACT

The past decade has witnessed many great successes of machine learning (ML) and deep learning (DL) applications in agricultural systems, including weed control, plant disease diagnosis, agricultural robotics, and precision livestock management. Despite tremendous progresses, one downside of such ML/DL models is that they generally rely on large-scale labeled datasets for training, and the performance of such models is strongly influenced by the size and quality of available labeled data samples. In addition, collecting, processing, and labeling such large-scale datasets is extremely costly and time-consuming, partially due to the rising cost in human labor. Therefore, developing label-efficient ML/DL methods for agricultural applications has received significant interests among researchers and practitioners. In fact, there are more than 50 papers on developing and applying deep-learning-based label-efficient techniques to address various agricultural problems since 2016, which motivates the authors to provide a timely and comprehensive review of recent label-efficient ML/DL methods in agricultural applications. To this end, we first develop a principled taxonomy to organize these methods according to the degree of supervision, including weak supervision (i.e., active learning and semi-/weakly- supervised learning), and no supervision (i.e., un-/self- supervised learning), supplemented by representative state-of-the-art label-efficient ML/DL methods. In addition, a systematic review of various agricultural applications exploiting these label-efficient algorithms, such as precision agriculture, plant phenotyping, and postharvest quality assessment, is presented. Finally, we discuss the current problems and challenges, as well as future research directions. A well-classified paper list can be accessed at https://github.com/DongChen06/Label-efficient-in-Agriculture.

Motivation & Objective

• Address the high cost and scarcity of large-scale labeled datasets in agricultural machine learning applications.
• Systematically categorize label-efficient learning (LEL) methods based on supervision levels: weak supervision (active, semi-, weakly supervised) and no supervision (unsupervised, self-supervised).
• Review state-of-the-art LEL techniques applied to key agricultural domains such as precision agriculture, plant phenotyping, and postharvest quality assessment.
• Identify open challenges in LEL for agriculture, including data distribution shifts, catastrophic forgetting, and semantic gaps in multi-modal learning.
• Provide a curated, up-to-date list of relevant papers via a GitHub repository to support ongoing research in label-efficient agricultural AI.

Proposed method

• Developed a principled taxonomy to classify LEL methods by supervision level: weak supervision (active learning, semi-/weakly-supervised learning) and no supervision (unsupervised/self-supervised learning).
• Reviewed representative state-of-the-art LEL techniques, including pseudo-labeling, consistency regularization, contrastive learning, and knowledge distillation.
• Evaluated the integration of LEL with incremental learning to handle evolving data streams and new concept learning without forgetting.
• Explored multi-modal learning from unlabeled data using visual, depth, and other sensor modalities to create self-supervision signals.
• Proposed strategies such as online clustering and mixture of Gaussians to dynamically update model representations in latent space.
• Analyzed the use of self-supervised pre-training on unlabeled agricultural data to improve downstream task generalization and robustness to distribution shifts.

Experimental results

Research questions

• RQ1 How can label-efficient learning methods reduce the dependency on large-scale, costly human-annotated datasets in agricultural machine learning?
• RQ2 What are the key technical differences and performance trade-offs among weakly supervised, self-supervised, and unsupervised learning in agricultural applications?
• RQ3 In what ways do label-efficient methods improve model generalization when labeled data is scarce or imbalanced?
• RQ4 How can incremental learning and continual representation updating mitigate catastrophic forgetting in dynamic agricultural environments?
• RQ5 What are the opportunities and challenges in applying multi-modal, label-efficient learning to agricultural tasks such as weed detection and fruit harvesting?

Key findings

• More than 50 research papers on label-efficient deep learning in agriculture have been published since 2016, indicating growing research interest.
• Self-supervised and semi-supervised learning methods show strong potential in reducing annotation costs while maintaining high performance in tasks like plant disease detection and weed recognition.
• Pseudo-labeling and consistency regularization techniques improve model robustness on unlabeled data, though they remain vulnerable to confirmation bias.
• Incremental learning strategies such as online clustering and adaptive Gaussian mixture models help maintain performance when new data or concepts emerge.
• Multi-modal learning from unlabeled data—using visual, depth, and other sensor inputs—can enhance feature representation and is underexplored in agriculture.
• The GitHub repository (https://github.com/DongChen06/Label-efficient-in-Agriculture) provides a well-classified, actively maintained list of relevant papers to support future research.