Advanced techniques and applications of LiDAR Place Recognition in Agricultural Environments: A Comprehensive Survey

An optimal solution to the localization problem is essential for developing autonomous robotic systems. Apart from autonomous vehicles, precision agriculture is one of the elds that can bene t most from these systems. Although LiDAR place recognition is a widely used technique in recent years to achieve accurate localization, it is mostly used in urban settings. However, the lack of distinctive features and the unstructured nature of agricultural environments make place recognition challenging. This work presents a comprehensive review of state-of-the-art the latest deep learning applications for agricultural environments and LPR techniques. We focus on the challenges that arise in these environments. We analyze the existing approaches, datasets, and metrics used to evaluate LPR system performance and discuss the limitations and future directions of research in this eld. This is the rst survey that focuses on LiDAR based localization in agricultural settings, with the aim of providing a thorough understanding and fostering further research in this specialized domain.

💡 Research Summary

This survey provides a comprehensive overview of LiDAR‑based place recognition (LPR) techniques as they apply to agricultural environments, a domain that has received far less attention than urban settings despite its growing importance for autonomous farming robots. The authors begin by motivating the need for precise localization in precision agriculture, noting the global population rise, aging farmer workforce, and the potential for robotics to reduce labor costs, increase yields, and lower environmental impact. They argue that reliable place recognition is a prerequisite for tasks such as autonomous vehicle navigation, path planning, loop‑closure in SLAM, and long‑term map maintenance, especially when GNSS signals are unreliable or subject to multipath effects in fields and orchards.

The paper then surveys recent deep‑learning applications in agriculture over the past five years, categorizing them into precision agriculture, autonomous agricultural vehicles, crop monitoring, and phenotyping. In each category, the authors highlight how LiDAR data complement or surpass visual sensors, particularly under varying illumination and weather conditions. For autonomous vehicles, examples such as DeepWay and contrastive‑learning‑based row clustering demonstrate how LPR can improve row‑following and waypoint generation in vineyards. In phenotyping, multi‑temporal LiDAR registration (e.g., the Pheno4D dataset) enables longitudinal analysis of plant growth and deformation.

A core contribution of the survey is the systematic analysis of LPR algorithms themselves. Traditional approaches—keypoint detectors (ISS, Harris3D) and global feature encoders (PointNet, DGCNN, VoxNet)—are shown to struggle in unstructured, seasonally changing fields where distinctive geometric cues are scarce. The authors therefore focus on three emerging families of methods that have shown promise for agriculture: (1) multi‑scale point‑net architectures that capture both local texture and global layout; (2) contrastive or self‑supervised learning frameworks that generate positive/negative pairs from unlabeled point clouds, reducing the need for costly annotation; and (3) temporal registration networks that explicitly model seasonal variations to maintain a consistent embedding over time.

The survey also compiles the few publicly available agricultural LiDAR datasets, including Pheno4D (maize and tomato growth sequences), the Agricultural LiDAR Dataset (ALD) with row and weed annotations, and a high‑resolution vineyard scan set. For each dataset the authors detail sensor specifications (channel count, scan rate), labeling granularity (row, crop type, growth stage), and typical usage scenarios (training, validation, domain adaptation).

Evaluation metrics are another focal point. While Recall@1/5 and mean Average Precision (mAP) remain standard, the authors argue that agricultural LPR requires additional measures of temporal stability and spatial accuracy. They propose a Temporal Consistency Score (TCS) to quantify how consistently the same physical location is recognized across different growth stages, and a Localization Drift metric (expressed in meters per second) to capture cumulative positional error during long‑duration missions.

Finally, the paper identifies current limitations and outlines future research directions. Key challenges include (i) severe point‑cloud noise caused by weather, foliage motion, and seasonal changes; (ii) the scarcity of large, richly annotated agricultural LiDAR datasets; (iii) high computational demands of state‑of‑the‑art deep models, which hinder real‑time deployment on edge devices. To address these, the authors recommend (a) developing lightweight edge‑AI models optimized for inference on embedded GPUs or specialized accelerators; (b) exploiting multimodal sensor fusion (LiDAR + RGB/D + GNSS) to improve robustness; (c) advancing semi‑supervised and domain‑adaptive learning to leverage synthetic or cross‑domain data; (d) creating simulation environments for data augmentation and algorithm testing; and (e) establishing standardized benchmarks and open‑source frameworks to promote reproducibility and industry‑academia collaboration.

In summary, this survey fills a critical gap by consolidating the state of the art in LiDAR‑based place recognition for agricultural settings, evaluating datasets and metrics, and charting a roadmap for future advances that will enable reliable, autonomous operation of farming robots across diverse, unstructured, and seasonally evolving fields.