This review synthesizes advancements in proximal spectral sensing devices—including portable, vehicle-based, UAV-based, and IoT-based—for monitoring field crop growth traits. By evaluating their technical capabilities, applications, and limitations, it addresses critical challenges in scalability, data integration, and environmental adaptability to advance precision agriculture (PA) practices.
A systematic analysis of literature (2001–2024) was conducted using keywords such as “proximal remote sensing,” “spectral sensors,” and “crop growth monitoring” in the Web of Science database, yielding 1,278 publications. The performance, sensing mechanisms, and practical applications of these devices were analyzed across platforms, with a focus on their ability to estimate key growth indicators (e.g., biomass, leaf area index, nitrogen content) and resolve PA-related challenges.
Portable spectral sensors excel in capturing high-resolution, targeted measurements but face limitations in accuracy during early crop growth stages and under complex field conditions. Vehicle-based systems enable efficient large-area scanning but encounter synchronization challenges between sensors and machinery, alongside susceptibility to environmental interference. UAV-based devices deliver rapid, high-throughput data collection but require enhanced endurance and integration with satellite imagery to achieve regional scalability. IoT-based networks support continuous monitoring but are constrained by a lack of specialized spectral sensors and insufficient durability in harsh agricultural environments. Cross-platform data fusion remains impeded by heterogeneity in data types, spatial scales, and storage protocols, while device durability, algorithmic robustness, and environmental resilience emerge as critical factors for reliable field deployment.
Proximal spectral sensing devices hold transformative potential for multi-scale crop growth monitoring, yet persistent technical gaps hinder their widespread adoption. Future research should prioritize the development of lightweight hyperspectral imaging systems paired with advanced computational algorithms, unified frameworks for cross-platform data fusion, and durable IoT sensors tailored for harsh field conditions. Additionally, integrating UAV-based data with satellite observations will enhance regional insights, while standardized protocols and interdisciplinary collaboration are essential to bridge ground-to-space monitoring networks. These advancements will foster intelligent, sustainable crop management systems, ultimately addressing global agricultural productivity and sustainability challenges.
Automated plot extraction in agronomic research field trials is essential for high-throughput phenotyping and precision agriculture. Accurate delineation of plot boundaries enables reliable crop type classification, yield estimation, and crop health monitoring. However, traditional plot extraction methods rely heavily on manual digitization, which is time-consuming, labor-intensive, and prone to inconsistencies. This study aims to develop a Segment Anything Model (SAM)-based framework that automates plot extraction while maintaining high accuracy across diverse agricultural field conditions.
The proposed framework consists of mask generation, plot orientation estimation, and plot refinement. SAM is leveraged to generate plot masks, which are subsequently filtered and refined to ensure precise boundary delineation. The method is designed to function without the need for model training or fine-tuning, making it highly adaptable across different datasets.
The framework was validated on five datasets, demonstrating robust performance under varying field conditions. The pixel-based evaluation yielded an average F1 score of 89.54%. For polygon-based evaluation, the framework achieved 99.71% precision at IoU=50% and an average precision of 68.51% across IoU thresholds from 50 to 95%, confirming its ability to accurately extract plot boundaries. A Canopeo-based regression analysis further demonstrated that the extracted plots provide more reliable phenotypic estimates compared to manually digitized ground reference data.
The proposed framework significantly reduces manual effort while ensuring high precision and scalability for large-scale phenotyping applications. By relying solely on RGB imagery and zero-shot segmentation, it enhances accessibility for real-world agricultural research. Future work will focus on extending the framework to irregular plot structures, diverse crop types, and computational optimizations for large-scale implementation.
The majority of newly developed sprayers now feature advanced capabilities, allowing herbicide application with centimeter-level precision, potentially reducing herbicide use by up to 90%. However, accurately identifying the precise locations to spray, known as the application map, remains a significant research challenge. Recently, both commercial providers and research institutions have proposed various drone-based methods for generating application maps. Despite these advancements, practical adoption is limited, primarily due to regulatory constraints and the high costs associated with the technology. A promising approach to increasing the adoption of these technologies lies in the utilization of more cost-effective hardware solutions. In this paper, we introduce and evaluate a novel detection method specifically designed for identifying Rumex obtusifolius (sorrel) and for automatically generating application maps that are compatible with most GNSS-enabled sprayers. To this end, we present a new metric for treatment success, termed the treatment F1-score, and conduct a comparative analysis of the performance of the DJI Mini 2 and the DJI Matrice 350 RTK using our proposed system, achieving treatment F1-scores of 0.61% and 0.65% , respectively. The ability of this system to deliver good performance utilizing significantly less expensive hardware than typically employed in similar applications suggests a potential for broader adoption, particularly given the unexpectedly modest performance gap of only 4 percentage points in the treatment F1-score. Under controlled experimental conditions, we observed reductions in herbicide use of up to 97% without missing any targets. In practical applications within real-world meadows, a 40% reduction in herbicide consumption was achieved with a treatment accuracy of 85% . These findings underscore the substantial potential for future technological advancements. The standalone object detector achieves a mean Average Precision (mAP) of 67.4% and an F1-score of 62%, demonstrating robust performance even on out-of-distribution drone data collected by other researchers. Still, the performance of the object detection algorithm is identified as a critical bottleneck in the system. To facilitate further research and development in this domain, we have made our training dataset available for download.
Spot spraying represents an efficient and sustainable method for reducing herbicide use in agriculture. Reliable differentiation between crops and weeds, including species-level classification, is essential for real-time application. This study compares state-of-the-art object detection models-YOLOv8, YOLOv9, YOLOv10, and RT-DETR-using 5611 images from 16 plant species. Two datasets were created, dataset 1 with training all 16 species individually and dataset 2 with grouping weeds into monocotyledonous weeds, dicotyledonous weeds, and three chosen crops. Results indicate that all models perform similarly, but YOLOv9s and YOLOv9e, exhibit strong recall (66.58 % and 72.36 %) and mAP50 (73.52 % and 79.86 %), and mAP50-95 (43.82 % and 47.00 %) in dataset 2. RT-DETR-l, excels in precision reaching 82.44 % (dataset 1) and 81.46 % (dataset 2) making it ideal for minimizing false positives. In dataset 2, YOLOv9c attains a precision of 84.76% for dicots and 78.22% recall for Zea mays L.. Inference times highlight smaller YOLO models (YOLOv8n, YOLOv9t, and YOLOv10n) as the fastest, reaching 7.64 ms (dataset 1) on an NVIDIA GeForce RTX 4090 GPU, with CPU inference times increasing significantly. These findings emphasize the trade-off between model size, accuracy, and hardware suitability for real-time agricultural applications.
Current approaches for monitoring soil organic matter (SOM) exhibit limitations in long-term predictive accuracy and data efficiency. This study aims to develop a remote sensing framework that integrating Landsat imagery and three modeling algorithms (PLSR, RF, Cubist) to address these challenges, reduce sampling workload, and enable large scale soil fertility assessments. Feature selection via Boruta and recursive feature elimination (RFE) was applied to optimize model performance, with PLSR identified astheoptimal algorithm. The framework utilized long-term Landsat imagery (2007–2021) and an inter-annual migration learning approach to map SOM dynamics. PLSR achieved cross-year SOM prediction (R2 = 0.51, RMSE = 3.97 g/kg), enabling accurate mapping of non-sample years with minimal field data and long-term imagery. Analysis of SOM trends revealed a decade-long decline in the study area, strongly correlated with land-use intensity. The proposed inter-annual migration learning method demonstrates that SOM dynamics can be efficiently tracked using sparse sampling and time-series remote sensing, offering a scalable tool for soil fertility management and precision agriculture.
Timely and accurate monitoring of plant nitrogen concentration (PNC) is vital for optimizing field N management. Hyperspectral indices are commonly used as a predictor for monitoring the PNC of crops, but individual spectral indices are often susceptible to cultivars and growth stages. Machine learning (ML) is a promising method for mining more spectral variables to assess the PNC of crops. To monitor the PNC of potatoes, therefore, this study extended previous work to further use hyperspectral optimized spectral indices (OSI) as input variables of ML, while, comparing with the ML models that used full-spectrum (FS), existing spectral indices (ESI) and sensitive spectral bands (SSB) as input variables, as well as simple regression model based on OSI alone. The partial least squares regression (PLSR), random forest (RF), support vector regression (SVR), and artificial neural network (ANN) models were calibrated using a dataset encompassing three cultivars and critical fertigation growth stages under three to six N levels. The calibrated ML models were evaluated using the datasets from independent experiments and two farmers´ fields. The OSI as an input variable in ML models showed superiority for predicting the potato PNC compared to FS, SSB, and ESI. The OSI-based RF model with an R2 of 0.79, RMSE of 0.27%, and RPD of 2.18 had higher accuracy for predicting potato PNC than other ML models. Comparing the simple optimized spectral indices regression model alone, the OSI-based RF model reduced RMSE by mitigating the effects of cultivars and growth stages on PNC prediction. The OSI-based RF model significantly contributes to optimum fertilization management based on actual potato N status during critical growth periods.
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：