Abstract
Advancements in agricultural technologies have increasingly emphasized technical innovations aimed at improving the predictability and reliability of agricultural outputs. These aspects encompass developments in agricultural machinery, automation technologies, biotechnology, and controlled environment farming systems. This article focuses on Remote Sensing (RS)-based approaches applied to agricultural yield estimation for both crops and plants. RS technologies offer enhanced precision and scalability, making them particularly effective for large-scale agricultural monitoring and analysis. A systematic classification of RS-based methodologies employed for crop yield estimation is presented in this study. These methodologies are categorized into: (i) Sensor-Based approaches, (ii) Platform-Based approaches, (iii) Analytical and Modeling-based methods, and (iv) Machine Learning (ML)-driven models. Based on findings reported across multiple studies, it is observed that Deep Learning (DL)-based architectures consistently achieve superior performance across key evaluation metrics, including accuracy, precision, recall, and F1-score. This performance advantage stems from their capacity to learn hierarchical representations, capture complex non-linear relationships, scale efficiently with large datasets, and reduce reliance on manual feature engineering. Following this classification, our article presents a comprehensive discussion of the limitations associated with these methodologies. These challenges are organized into four major categories: (i) Environmental, (ii) Algorithmic, (iii) Hardware and Operational, and (iv) Wireless Sensor Networks (WSNs) related limitations. The adopted classification framework helps readers identify and address the key challenges associated with effective yield estimation in crops and plants. Moreover, the article concludes by outlining several future research directions intended to support and guide both early-career and experienced researchers in this domain.