Crop Yield Estimation Using Machine Learning and Deep Learning Techniques: A Survey with Special Focus on Fruit Yield Estimation in Precision Agriculture

Abstract

Crop yield estimation is a central task in precision agriculture because reliable estimates support crop management, harvest planning, storage, marketing, logistics, and food-supply decisions. Recent advances in sensing, remote sensing, computer vision, machine learning (ML), and deep learning (DL) have shifted yield estimation from predominantly manual or area-based procedures toward data-driven prediction and image-based estimation. This survey reviews the ML- and DL-oriented crop-yield estimation approaches represented in the 83 reference articles and studies in this paper, with particular attention to remote-sensing-based crop assessment and vision-based fruit yield estimation. The review organizes the literature according to sensing modality, learning paradigm, estimation target, and deployment requirement. Satellite and airborne imagery, multispectral and hyperspectral observations, LiDAR, thermal sensing, UAV imagery, ground sensors, and IoT systems are considered as complementary sources of information. The ML/DL discussion covers feature-based learning, neural-network regression, convolutional neural networks, object detection, segmentation, fruit counting, and lightweight real-time detectors. A dedicated discussion section synthesizes only findings reported in the cited studies; The survey shows that the strongest evidence concerns remote sensing for crop growth/yield variability and deep-learning-based fruit detection and counting. It also identifies persistent issues involving occlusion, illumination and environmental variation, annotation cost, cross-season generalization, sensor fusion, computational constraints, and the distinction between fruit detection accuracy and actual yield prediction. Finally, research directions are proposed around multimodal learning, temporal modeling, transfer learning, edge AI, uncertainty-aware prediction, and long-term field validation.