Proceedings
Topics
| Filter results62 paper(s) found. |
|---|
1. Enhancing the Reliability of Portable Soil Probes Through Machine Learning OptimizationPrecision agriculture requires high resolution soil data, yet traditional laboratory analyses limit sampling because of high human labour and analysis costs. Portable spectrographic tools allow rapid infield soil characterization with multiple measurements, although their accuracy often falls short of standard laboratory protocols. We hypothesized that machine learning (ML) models could improve the reliability of these tools by adjusting their outputs against laboratory reference values. ... A. Cambouris, E. Lord, M. Duchemin |
2. Influence of the Type of ANN Algorithm on Prediction of Georeferenced Sugarcane QualityOptimizing crop quality and yield is critical to the development of more sustainable agriculture on large scale. Predictive models can provide assessment of those attributes prior to harvesting using techniques of artificial intelligence to support site-specific management. The objective was to investigate the influence of ANN (Artificial Neural Network) algorithms on prediction of sugarcane quality. Brix content of sugarcane, variety CTC 2994 on second ratoon, was measured in laboratory usin... T. Canata, F.P. Monachesi, O.D. Salomão, L. Rodrigues , L.E. Zonfrilli, H.C. De La Cruz, P.F. De Mello |
3. A Counterfactual Modelling Framework with On-farm Experimentation for Guiding Site-specific Nitrogen ApplicationsNitrogen (N) fertiliser is a key driver of wheat grain protein content (GPC) and yield, and represents one of the largest variable input costs and sources of emissions in Australian grain production. Yet estimating optimal N fertiliser rates remains challenging due to spatio-temporal variability in soil N supply and crop nutrient demand, as well as dynamic interactions between yield, GPC, and water availability. On-farm experimentation (OFE) provides valuable insights into crop resp... M.J. Tilse, T. Bishop, S. Poole, P. Filippi |
4. A Multimodal Spectral-Robustness-LLM Pipeline for Non-Destructive Identification of Loropetalum chinense CultivarsProprietary cultivars of ornamental shrub Loropetalum chinense, particularly the visually and spectrally similar ‘Cerise Charm’, ‘Purple Daybreak’, and ‘Red Diamond’, derive their market value from the intensity and stability of anthocyanin pigmentation, a trait that degrades subtly under abiotic stress. Reliance on manual (visual) grading makes the industry vulnerable to these latent, pre-manifestation pigment losses, which are often detected only... P. Sundaravadivel, S. Borah, H. Manjunatha, S.P. Kumpatla, L. Tamil, P. Knight, T. Stroud |
5. Unified Detection and Weight Estimation of Small Fruits Using Multi-Task Vision Models in Precision AgricultureThis work presents a single computer vision model that can perform both object detection and image-level regression from the same input image. Many real applications, especially in agriculture, need information about individual objects as well as a global measurement for the entire image. When analyzing an image of small fruits such as different types of berries, grapes, currants, and muscadine grapes, it may be necessary to detect and classi... P. Sundaravadivel, T. Stroud, S. Borah, B.J. Sampson, P. Knight, S.P. Kumpatla, J.F. Ross |
6. Application of Machine Learning Algorithms and Remote Sensing for Predicting Losses in Peanut HarvestingPeanut (Arachis hypogaea L.) is a crop of substantial economic and social relevance in Brazil, particularly in the state of São Paulo, which accounts for the majority of national production and consistently attains high productivity levels. Despite significant advances in agricultural mechanization, harvesting remains one of the most critical phases of peanut production, especially during mechanical digging, a stage in which considerable yield losses frequently occur. These losses are ... G. Pereira Costa, A.L. Brito Filho, T.C. Oliveira, J. , R.P. Silva |
7. Automated Leak Classification in Drip Irrigation Systems using Deep Learning and RGB CamerasThe increasing demand for water efficiency in agriculture has driven the development of intelligent irrigation systems. Among them, drip irrigation is widely adopted due to its efficiency; however, these systems are susceptible to leaks caused by mechanical wear, animal interference, and adverse environmental conditions. The manual detection of leaks by human workers in drip irrigation systems is a time-consuming task, difficult to scale, ... F.P. Rivera, C. Kamienski |
8. Combining YOLOv9 and Fuzzy Inference System to Improve the Precision of Weed Recognition Systems in Soybean Crops Using UAV ImageryWeeds are a problem in crops because they compete with crops for nutrients, sunlight, and water, hindering their full development. To control these plants, herbicides are usually applied throughout the field. Therefore, to optimize the application process, many researchers have been working on automatic weed recognition systems based on artificial intelligence techniques for field imaging, enabling the localized application of herbicides. To this end, the YOLO (You Only Look Once) object dete... M. Tosin, I. Scheeren, C. Markus |
9. Impact of Telemetry Data Preprocessing on the Accuracy of Fuel Consumption Predictive Models in Heavy-Duty Truck Transport of Sugarcane StalksFuel consumption efficiency in biomass transport is a determinant factor for the sustainability of agribusiness. However, agricultural machinery telemetry data present intrinsic challenges, such as onboard sensor noise and inconsistencies. This study aimed to demonstrate that meticulous data preprocessing is more relevant than algorithmic complexity in achieving high-performance predictive models. The raw dataset contained 43,112 trips by trucks responsible for transporting sugarcane stalks f... L.E. Zonfrilli, A.R. Camolesi, T. Canata |
10. How Does Yield Data Filtering in Grain Harvesters Influence the Quality of Interpolated Maps?Yield maps generated from grain harvester data are effective tools for characterizing the spatial variability of crop yields. However, several embedded errors are inherent in these datasets, requiring removal methods to ensure the fidelity of actual field yield values and the reliability of the resulting maps. Therefore, this study aimed to determine the optimal combination of parameters for the global and local filtering of grain harvester yield maps to improve the quality of interpolated ma... A. Andrade Da Silva, E. Sales, T. Moura Oliveira, R. De Souza Silva, S. Luns, R.P. Silva, E. Perussi |
11. ICICLE: A Generic Cyberinfrastructure Pipeline for AI-Driven Digital Agriculture ProcessingDigital agriculture suffers from fragmented data and processing tools, restricting our ability to derive consistent, scalable insights that support precision management. The NSF ICICLE (Intelligent Cyberinfrastructure with Computational Learning in the Environment) project addresses this challenge by developing a generalized cyberinfrastructure framework for AI‑enabled data processing across diverse production systems. Deployed at The Ohio State University, ICICLE provides a unif... H. Subramoni, S.A. Shearer, J.P. Fulton |
12. Integrating Management Zones, Artificial Neural Networks and Remote Sensing for Smart Peanut HarvestingThe integration of technologies contributes significantly to agricultural development, especially regarding the rational and more sustainable use of soil. Thus, the use of remote sensing and artificial intelligence techniques combined with precision agriculture can maximize smart harvesting for peanut crops, which face several challenges such as limited harvesting technology, indeterminate growth, and the development of pods below the soil surface. Therefore, this study aimed to develop a pea... |
13. Alternative Method for Measuring Fuel Consumption in Agricultural Machinery Using Arduino and Flow SensorsMonitoring fuel consumption in agricultural machinery is a strategic component of precision agriculture, as it is directly associated with operational efficiency, cost reduction, and the mitigation of CO₂ emissions. Despite technological advances in agricultural tractors, most machines, including recent models, do not feature dedicated sensors for direct fuel flow measurement, relying instead on visual fuel level indicators or estimates based on engine parameters, which limits the accuracy ... R. De Souza Silva, L.E. Zonfrilli, A. Andrade Da Silva, R.P. Silva, A.D. Carreira |
14. Accessible Spectral Engineering: PLSR Optimized for Organic Carbon EstimationIn the context of Agriculture 4.0, monitoring Soil Organic Carbon (SOC) is crucial for precision agriculture and sustainability, as it is the determining variable for water retention capacity, chemical fertility, and atmospheric carbon sequestration. While the global technological frontier is advancing toward sensors in the Thermal Infrared (TIR) spectrum, adoption in Latin America faces structural barriers of infrastructure and cost. In Peru, the official Walkley-Black method, although norma... A. Isla Aguilar |
15. A Real-Time Intelligent Framework for Wheat Stripe Rust Management Using Lightweight Deep Learning and LLMsWheat stripe rust (Puccinia striiformis f. sp. tritici) poses a severe threat to global food security, necessitating rapid and precise disease grading for site-specific intervention. While edge-computing devices offer on-site monitoring potential, balancing real-time accuracy with cognitive decision-making remains a challenge. This study proposes an integrated intelligence framework that synergistically fuses lightweight visual perception with Large Language Model (LLM)-driven cognitive reaso... W. Su, W. Su, W. Su, W. Su, W. Su |
16. Artificial Intelligence for Management Zone Delineation: A Bibliometric Review (2008-2025)This bibliometric review aims to map research trends, key terms, and leading institutions in the use of artificial intelligence (AI) methods, with emphasis on Machine Learning (ML), Deep Learning (DL), and Neural Networks, applied to the delineation of management zones (MZs) in Precision Agriculture (PA). The analysis was conducted using the Scopus database, applying a structured search string to titles, abstracts, and keywords. Metadata were collected on September 26, 2025. The initial searc... L. De Goes Sterle, J.P. Molin, R. Fray Da Silva |
17. Precisely Monitoring Nitrogen Requirements for Winter Wheat and Spring Barley Based on Crop Yield Predictions, Remote Sensing Imagery and Soil Texture MapsThe timely and precise evaluation of crop nitrogen demand is crucial for maximizing farmers' contribution margin while simultaneously minimizing nitrogen fertilization. Sufficient nitrogen fertilizer has to be provided for adequate crop growth, yet fertilization should not be excessive to ensure its environmental impact is minimized. A variety of factors determine nitrogen demands as well as crop yield, including weather, topography and soil texture. These factors vary spatially, meaning ... N. Hollain, S. Wang, B. Feld Mikkelsen, |
18. An Interpretable Machine Learning Framework for Soil Nutrient Assessment Based on pH and Electrical ConductivityUnderstanding how the physical and chemical properties of soil influence nutrient availability is fundamental for advancing precision agriculture, as these properties directly affect the efficiency of macro- and micronutrient absorption by plants. In recent years, the increasing availability of open agricultural datasets has created new opportunities for developing data-driven frameworks capable of supporting large-scale soil assessment and decision-making. However, the effective integration ... T.A. Rodolfo, O.J. Gonzalez Zarate, C. Gonzalez Aguilera |
19. Deep Learning Models Applied to Drone Imagery for Counting, Biometry, and Carbon Stock Estimation in Large-scale Macaw Palm (Acrocomia Aculeata) PlantationsMacaw palm is a native Brazilian species with significant productive potential, emerging as a premier candidate for the sustainable replacement of oil palm and as a strategic feedstock for sustainable aviation fuel (SAF) and carbon credit markets. However, as the crop is still in the process of domestication and commercial expansion, there is an urgent need for efficient monitoring technologies that enable the identification of superior individuals and the rigorous auditing of carbon stocks a... P.M. De Sousa, V.A. Galvan, J. Souza, R. . De Oliveira , L.D. Corrêdo, L.D. Pimentel, B.C. Albuquerque |
20. Evaluation of Convolutional Neural Network Architectures for Stress Detection in Eucalyptus saligna Using Multispectral UAV ImageryRoot malformation disorder (RMD) is a significant abiotic condition that impairs water and nutrient uptake in Eucalyptus saligna, leading to physiological stress and reduced stand uniformity. Early detection in commercial plantations is hampered by the extensive spatial scale and logistical limitations of manual field inspections. In this context, the present study evaluated the performance of three convolutional neural network (CNN) architectures, U-Net, U-Net++, and Attention U-Net, in dete... L.D. Amaral, S.D. Pereira Da Silva, R.A. Fantinel, V. Richter, N. |
21. Integrating Data Layers with Machine Learning to Predict Yield for Irrigated Grain Crops within Management ZonesEffective yield prediction is fundamental for precision agriculture, enabling data-driven management decisions. This study, conducted in a 52.3 ha center-pivot irrigated field in Itaí, São Paulo, Brazil, evaluated the hypothesis that delineating management zones (MZs) based on stable soil and terrain attributes, combined with machine learning (ML) algorithms, improves grain yield prediction accuracy compared to field-scale models. Apparent soil electrical conductivity (ECa) at t... L. De Goes Sterle, R. Canal Filho, V. Ferraz, M. Gelain, J.P. Molin |
22. Detection of Maize Foliar Diseases Using AI Optimized for Deployment on Edge DevicesMaize is a strategic crop for both regional and global food security. Its productivity is significantly affected by several foliar diseases, among which—common rust, gray leaf spot, and blight—are some of the most prevalent and damaging. These pathologies can cause substantial yield losses if not detected and treated in a timely manner, making early diagnosis a fundamental factor to ensure healthy and sustainable crop development. However, traditional diagnostic methods based on m... O.J. González Zarate, L. Macea Zabaleta, N. Castillo Ojeda, A.F. Flórez Olivera, T.A. Rodolfo, C. Gonzalez Aguilera |
23. Evaluating Deep Learning Models for Image-Based Corn Kernel Detection, Counting and Yield PredictionAccurate estimation of kernel number in corn is essential for evaluating yield potential in breeding and agronomic research. However, manual kernel counting is labor-intensive, prone to human error, and impractical for large-scale datasets, while most existing automated devices are limited to simple counting tasks. This study evaluates deep learning-based approaches for automated kernel detection and counting using You Only Look Once models and Faster R-CNN. Specifically, YOLOv8x, YOLOv10x, a... B. Ghimire, L. Lacerda, T. Bourlai, G. Lu |
24. Artificial Intelligence Framework for Bioenergetic Flow Audit and Thermal Entropy: A Precision Approach in the Brazilian SemiaridExtensive livestock farming in the Brazilian semiarid faces productivity bottlenecks masked by weed competition, where invasive plants mimic vegetative vigor but impose thermal and nutritional stress on the herd. This study aimed to develop and validate the "Predict-IA" framework, a bioenergetic audit tool based on Artificial Intelligence to quantify systemic entropy and "energy leakage" in degraded pastures. A ten-year time series (2015-2025) of Sentinel-2 multispectral d... A. Alkmim, D. Vitor Dos Santos |
25. Study on the Phenological Zoning Method for Winter Wheat in the Huang-Huai-Hai Region of ChinaThe impact of global climate change on agricultural phenology is becoming increasingly significant. As a major producer of winter wheat, China's cultivation areas span multiple climate zones. Against the backdrop of climate change, the spatiotemporal differentiation of crop phenology has raised new scientific demands for agricultural zoning. Phenological zoning has guiding significance for variety selection, irrigation management, and pest prediction. However, existing research often reli... S. Xiaoyu, Q. Wu, Y. Ma, J. Zhang, P. Dong, X. Xu |
26. Plot2Phenome: A UAV-Based Deep Learning Framework for Automated Micro-Plot Segmentation and Plot Level PhenotypingAutomated micro-plot segmentation is a foundational requirement for plot-level phenotyping from UAV orthomosaics in field breeding trials. However, reliable delineation of individual plots remains difficult in realistic agronomic settings, particularly under canopy closure that erodes inter-plot gaps and under irregular or degraded plot boundaries caused by lodging, variable emergence, and field operations. These conditions reduce boundary contrast, increase instance adjacency, and introduce ... Y. Li, C. Jin, X. Zhang |
27. Machine Learning and Causal Analysis to Support Improved Crop Decision-makingWhile machine learning (ML) models, particularly Extreme Gradient Boosting (XGBoost) and Random Forest (RF), have demonstrated potential in generating accurate crop yield predictions, their practical adoption for on-farm decision support remains limited. A key challenge lies in their fundamentally associative nature, which, without additional tools, can reduce interpretability and diminish practitioner confidence. Explainable Artificial Intelligence (XAI) techniques like SHAP values address o... M. Chan Fu Wei, J.P. Molin, A. Colaço, L. Longchamps |
28. Herbicide Savings and Weed Control Performance Using Green-on-Green Spot Spraying in SoybeanThe conventional approach to weed control in large-scale soybean production relies on full-area herbicide spraying, resulting in high chemical input and operational costs. In this context, artificial intelligence-based spot spraying has emerged as a promising alternative to increase efficiency and reduce environmental impact. This study evaluated the performance of a green-on-green spot spraying system based on deep learning algorithms, CORTEX AI (Soybean Model v08), for post-emergence weed c... D. Gabriel, A.S. Fiegenbaum, S.R. Griebeler, M. Franchi, N.S. Dos Santos, K.F. Rocha, A.L. Vian |
29. Comparative Evaluation of Combined and Task Specific Detectors for Pomegranate Yield and Fruit Loss DetectionFruit cracking and drop represent major sources of yield loss in pomegranate orchards; however, existing vision-based yield estimation methods focus on counting healthy fruit and do not usually capture losses occurring on-tree and on the orchard floor, thereby constraining their operational relevance. This study evaluates detection strategies for simultaneous yield and loss quantification, with a specific comparison between combined multi class models and task specific single class models.... Y. Tenenboim, Y. Edan, I. Ginzberg, T. Paz Kagan |
30. Cross-Season Transfer Learning for Prawn Morphometric Estimation Using YOLOv11-PoseThe problem of maintaining accurate computer vision models in dynamic aquaculture pond environments is increasingly important as real world imaging conditions vary over time. Even in controlled indoor ponds, factors such as water turbidity, lighting angle, background reflections, and camera setup can change between monitoring sessions or seasons. These variations introduce domain shifts that can significantly degrade the performance of deep learning models trained under controlled conditions.... T. Carmon, E. Aflalo , A. Sagi, Y. Edan |
31. Universal Dataset Constructor & Preprocessing Framework for Earth Observation AI Tasks in Digital AgricultureThe rapid advancement of Artificial Intelligence (AI) in digital agriculture is increasingly dependent on the ability to fuse heterogeneous data sources. While Earth Observation (EO) data from Sentinel and Landsat missions provides a backbone for monitoring, high-performance models for yield prediction and land management require a more holistic approach. This paper presents a Universal Dataset Constructor & Preprocessing Framework designed to automate the generation of combined, multimod... V. Sorokina, I. Klinkov, D. Yablonski, S. Henkler, A. Zakhary |
32. Management Zone Delineation for the Optimization of Nitrogen Use Efficiency in Arabica Coffee CropsPrecision coffee farming requires efficient methods for Nitrogen (N) management—an input of high cost and environmental impact, whose optimization faces challenges due to the topographical characteristics of regions such as the Zona da Mata in Minas Gerais, Brazil. This study evaluates and compares different dimensionality reduction models for agricultural management zone (MZ) delineation, aiming to maximize nitrogen fertilizer use efficiency in Arabica coffee plantations. A dataset com... D.N. Nunes, R.P. Oliveira, L.D. Corrêdo, L. Peternelli, A.W. Pedrosa, V.H. Galvan, J. Souza |
33. Comparative Analysis of YOLOv3–YOLOv12 Architectures for Automatic Oil Palm Detection in Agricultural MonitoringOil palm (Elaeis guineensis) is considered the most productive oilseed crop worldwide, and Brazil holds one of the greatest global potentials for palm oil production. Efficient monitoring of cultivated areas is therefore essential for proper crop management, enabling the detection of planting gaps, yield estimation, and decision-making support. In this context, computer vision techniques based on deep learning models, particularly those from the YOLO (You Only Look Once) family, have... M.C. Arnosti, A. Felipe Dos Santos, T. Costa Barboza, L.S. Souza Pinto, E. Amaral, G. Lacerda Da Silveira, G. Valdes Fernandez |
34. Hardware–Software Co-Design of Quantized CNN Inference for Edge AI in Precision AgriculturePrecision agriculture increasingly relies on real-time automated inspection systems to ensure crop quality and reduce manual labor in grain handling processes. Manual visual inspection, traditionally used for grain quality assessment, is inherently limited by low throughput, subjectivity, and high labor costs. To address these issues, automated vision-based inspection systems have been widely adopted in industrial environments, enabling high-throughput and consistent grain classification. Rec... E. Marañon Aguilar, F. Kastensmidt, F. Benevenuti, C. Gonzalez Aguilera |
35. Automated Detection of Melons (Cucumis melo L.) via Multispectral UAV and Deep Learning in HondurasAccurate agricultural production estimation is vital for the logistical and financial efficiency of agribusiness. This study proposes an automated melon detection pipeline using a multispectral Unmanned Aerial Vehicle (UAV) and deep learning architectures. The experiment was conducted in Apacilagua, Honduras, during the 2023–2024 season, covering an area of 32.17 ha. Data collection took place between 90 and 100 days after sowing (DAS)—a critical maturation phase—using a DJI... E. Torres Avila, C.L. Bazzi, S. Moro Lumertz, E. Cely Bonilla, M. Furlan Maggi, T.A. Medeiros, D. Perez, K. Schenatto, R. Sobjak |
36. Soil Texture Classification by Image: Deep Feature Learning vs. Handcrafted Methods for Precision AgricultureAccurate soil texture classification is fundamental for precision agriculture, as it enables site-specific crop management that optimizes the utilization of agricultural resources and enhances overall crop productivity. This study presents a comparative analysis between features automatically extracted by a pre-trained SqueezeNet convolutional neural network (CNN) and three classical methods for manual feature extraction: Fast Fourier Transform (FFT), Gabor Filters, and Local Binary Patterns ... J.R. Favan , G.D. Faulin, F.M. Kasita Kashima, J. . Alegre, L.S. Gonçalves |
37. Enhancing Pest Detection Through Spectral Signature Extraction in Hyperspectral DataDetecting insect infestation is essential for effective crop protection, particularly in large-scale systems. Caterpillars and stink bugs induce physiological and structural alterations in plant tissues that can be captured through hyperspectral reflectance sensing, which is a non-destructive technique that measures plant responses across hundreds of wavelengths. However, raw spectral signatures are characterized by high dimensionality, strong inter-band correlation, and they often exhibit ba... A.O. Françani, J. Ferreira , L. Zhao, L.A. Jorge, K.M. De Oliveira, J.C. Felipe |
38. Estimation of Leaf Chlorophyll Index in Corn Using Smartphone Images and Machine LearningAccurate estimation of the Leaf Chlorophyll Index (LCI) in corn is fundamental for nitrogen management in precision agriculture, as nitrogen availability directly affects chlorophyll production and photosynthetic capacity. Conventional field assessment techniques are time-consuming and labor-intensive, while smartphone use provides a practical and low-cost alternative for obtaining high-resolution data in near-real-time. The objective of this study was to develop and validate a non-destructiv... C.C. Santana, S.M. Mintesinot, D. Queiroz, F.S. Santos, A.L. Coelho |
39. Autonomous Mobile Robot for Monitoring and Control of the Cotton Boll WeevilThe cotton boll weevil (Anthonomus grandis Boheman) is the key pest of Brazilian cotton, accounting for about 12% of production costs and, together with yield losses, reaching roughly R$ 2,470 per hectare per season. Because its immature stages develop protected inside the plant, insecticides reach only the adult, which entrenches calendar-based broadcast spraying and an average of 18 full-field applications per season. Conventional scouting samples as few as 0.1 points per hectare, leav... R. Balducci Borges, N. Smaal, P.V. Santos, W.G. Bendinelli |
40. Deep Learning in Seed Vigor Assessment: Analysis of U-Net Family Architectures for Soybean Seedling SegmentationIn response to the growing challenges faced by agricultural production, increasing productivity in already cultivated areas has become essential to ensure global food security. In this context, the use of high-vigor seeds is crucial to achieving higher crop yields. However, traditional vigor assessment methods are time-consuming, often manual, and dependent on specialized labor, which drives the search for automated solutions based on Computer Vision and Deep Learning techniques. Several stud... J. Martins Neto, P. Dos Santos E Silva, E. Freitas, D. G. Gomes, H.F. Abud |
41. Segmentation of Morphological Structures of Soybean Seedlings Using Convolutional Neural NetworksSoybean is one of the most relevant and highly demanded agricultural commodities worldwide, playing a central role in the production chains of food, animal feed, and vegetable oil. Globally, there is a growing demand for food, which has driven strategies to meet the needs of the world population. Many of these strategies involve expanding arable land and excessively exploiting soil nutrients in an unsustainable manner. An alternative strategy to address this issue is to incorporate seed lot q... P. Dos Santos E Silva, J. Martins Neto, E. Freitas, D. G. Gomes, H.F. Abud |
42. UAV-Based Multispectral Modelling of Biomass and Crude Protein Yield for Green Biorefinery ApplicationsIn animal production systems, protein demand is steadily increasing due to global population growth. This rising demand has highlighted the need to identify alternative and sustainable protein sources. Green biorefinery systems can efficiently extract protein from plant biomass. Previous studies confirmed that perennial grass crops such as Perennial Ryegrass, Festulolium, and Tall Fescue can produce high-quality biomass suitable for protein extraction. An estimation model of biomass yield and... M. Canciani, E. Han, U. Jørgensen, Y. Ikeda, N.P. Hansen, S.K. Jensen, M.R. Weisbjerg, T. Didion |
43. Unlocking Partially Annotated Agricultural Data: A Cut-and-Paste Data Augmentation Framework for Plant DetectionAccurate weed and crop recognition is essential for effective management practices in precision agriculture, enabling targeted herbicide application through automated spraying systems. However, the performance of deep learning models in real-world field settings is often limited by class imbalance, where broad categories such as monocotyledons and dicotyledons overshadow classes labelled at the species level. A major bottleneck in addressing this imbalance is the massive under-utilization of ... M. Madsen, , R.N. Jørgensen |
44. From Render to Field: Detecting Asian Soybean Rust Using Models Trained Exclusively on Synthetic ImageryTraining machine learning models for crop disease detection requires large, annotated datasets that are costly and difficult to obtain under variable field conditions. Asian Soybean Rust (ASR) can reduce soybean yield by up to 90% and costs Brazilian producers over US$2 billion per season in fungicide applications and yield losses. Despite this impact, existing machine learning studies on ASR remain scarce with no publicly ... L.B. Fontoura, A. De Freitas, E. Farinati Leite, J. Valiati |
45. Estimating Peanut Losses Using Machine Learning with Soil and Weather DataMechanized peanut harvesting is an important phase of the production system, directly affecting both production costs and crop yield. However, due to interactions among soil conditions, plant characteristics, and machine performance, this operation is carried out under challenging conditions that may result in high levels of loss. These losses are classified as visible when pods remain on the soil surface after digging and as invisible when they are incorporated into the soil profile, making ... A. Lopes De Brito Filho, F. Morlin Carneiro, G. Pereira Costa, B. Dos Santos Silva, P.H. Nogueira Gusmão, R.P. Pereira Da Silva |
46. Temporal Stability of Management Zones Derived from Vegetation Indices and Yield Data in Contrasting Production SystemsThe delineation of management zones is a central component of site-specific crop management in precision agriculture. However, the temporal stability of zones derived from different data sources remains a key challenge, particularly when vegetation indices and yield data are combined across multiple seasons. This study evaluates the temporal stability of management zones delineated using vegetation indices and yield data derived from long-term commercial field datasets. The proposed meth... |
47. Statistical Mean Comparisons in Unreplicated Yield Trials with Georeferenced DataPrecision agriculture technologies have enabled the collection of large volumes of georeferenced yield data within experimental fields. In practice, many on-farm experiments (OFE) are implemented as large unreplicated strips or field zones containing numerous observations within each zone. The lack of replication prevents the use of classical statistical models for comparing zone means. Although many yield observations are available per zone, spatial autocorrelation violates independence assu... M. Córdoba, P. Paccioretti, M. Balzarini |
48. Evaluation of Filtering Approaches and Spatial Relationships with Gaps in Sugarcane Yield MapsYield maps represent indispensable tools in Precision Agriculture for the quantitative and qualitative characterization of crops. However, data derived from yield monitors do not always reflect actual field yield, as they are subject to measurement errors, operational errors, and/or equipment malfunctions. The presence of such distorted information compromises analytical accuracy and, consequently, decision-making. Given this challenge, the present study aimed to evaluate the performance of f... Y. De Lacerda Barbosa, V. Ferraz, E. Rafael Otavio Da Silva, J.P. Molin |
49. Performance of Spatial Prediction Models Under Different Sample Densities in Corn Yield MapsYield maps are essential for the consistent management of crop variability. The accuracy of these maps, however, is directly affected by the density of data collected by grain harvesters and by the choice of interpolation method, whether deterministic or statistical. This study aimed to evaluate the performance of four spatial prediction methods, tested under different sample densities, for the development of corn yield maps. The study area corresponds to an 11-hectare commercial field, utili... L.M. Gimenez, L.M. Santos, M. Hass Bomfim Vieira |
50. Evaluation of the Performance of Computer Vision Models in the Detection and Counting of Tomato Plants Infected by Tomato Spotted Wilt Virus (TSWV)Tomato is one of the most economically important vegetable crops worldwide. However, this crop is severely affected by Tomato Spotted Wilt Virus (TSWV), whose transmission occurs mainly through thrips. Thus, identifying infected plants is an important step to reduce the dissemination and infection of healthy plants, reducing economic losses. Computer vision-based models have been widely used in the automated detection of plant diseases. In this context, this work aimed to evaluate the perform... L.S. Souza Pinto, S. . Azevedo, M. . Medeiros, A. Felipe Dos Santos, T. Costa Barboza, M.C. Arnosti, G. Valdes Fernandez , G. Lacerda Da Silveira |
51. Evaluation of Transfer Learning in Semantic Segmentation Models for Soybean SeedlingsSeed vigor evaluation is fundamental in the quality control of commercial lots, as it is directly associated with the rapid and uniform emergence of seedlings and the initial performance of crops in the field. Traditional methods, although widely used, present limitations such as long execution time, dependence on the evaluator’s experience, and subjectivity. In this context, systems based on Computer Vision emerge as promising alternatives for automating vigor assessment, as they enabl... E. Freitas, J. Martins Neto, P. Dos Santos E Silva, H.F. Abud, D.G. Gomes, V.C. Secundino |
52. Application of CNNs in Cattle Counting using RPAsThe increasing demand for productive efficiency and sustainability in the agricultural sector has driven the adoption of technologies focused on Precision Livestock Farming. Among the main operational challenges in extensive systems, the counting and monitoring of cattle herds stand out. Historically performed manually, these activities are time-consuming, increase labor costs, and are highly susceptible to human error, especially across vast territorial expanses. However, the parallel advanc... E. De Souza Salles, C. Souza, R. Clemente Thom De Souza |
53. Edge AI–Driven Soil Sensing and Fertilization Prediction for Solanum betaceumThe growing demand for data-driven fertilization strategies in high-value perennial crops has fostered the development of intelligent systems capable of supporting decision-making directly in the field. In the case of tree tomato (Solanum betaceum), fertilization is commonly performed based on fixed schedules or empirical criteria, which often fail to account for soil dynamics and nutrient variability. This study investigates the feasibility of an embedded artificial intelligence system that ... |
54. Improving In-season Corn Nitrogen Status Prediction using Satellite Remote Sensing and Foundation Models with Agronomic ConstraintsPrecision nitrogen (N) management in maize requires in-season estimates of crop nitrogen status that are both accurate and physiologically credible, yet agronomic training data are often limited because destructive sampling is expensive and spatially sparse. Mechanistic crop models respect physiology but require extensive calibration and are computationally costly at field scale, whereas purely data-driven machine learning using remote sensing can achieve good accuracy while producing implaus... A. Kechchour, Y. Miao, V. Sharma, D. Mulla |
55. Evaluation of Lettuce Image Classification with CNNs under Different NPK Nutritional ConditionsThe growing global demand for food has driven the development of technologies aimed at increasing productive efficiency in sustainable agricultural systems, such as hydroponics. In this context, proper monitoring of nutrient solutions is essential, particularly for the early detection of nitrogen (N), phosphorus (P), and potassium (K) deficiencies, which directly affect lettuce growth, yield, and quality. Traditional nutritional diagnostic methods often rely on destructive laboratory analyses... E.L. Silva, E. Freitas, V.C. Secundino, D.G. Gomes |
56. Semantic Segmentation Comparison of Prata Catarina Banana Bunches Using Convolutional Neural Network ModelsThe identification and classification of banana ripening stages are essential for production assessment in large scale plantations, enabling efficient harvest monitoring and ensuring fruit quality for commercialization. This study presents a comparative evaluation of three deep learning architectures applied to the semantic segmentation of Prata Catarina banana bunches, aiming to support automated monitoring systems and decision-making tools for precision agriculture applications. The evaluat... J. Rodrigues Moreno, E.L. Silva, E. Freitas, D. G. Gomes, Y. Costa G. Da Silva |
57. Detection of Banana Bunches and Peduncles in the Prata Catarina Cultivar Using Faster R-CNN With Transfer LearningBanana is one of the most produced and consumed fruits worldwide, being strategic for precision agriculture, especially in applications aimed at intelligent management and automated harvesting. Its economic and social relevance in tropical countries reinforces the need for technological solutions that increase productive efficiency and reduce losses in the field. In this context, the automatic detection of bunches and stalks in a natural environment represents a relevant challenge due to occl... Y. Costa G. Da Silva, E. Freitas, P.S. Costa, D.V. Beserra, D.G. Gomes |
58. Evaluation of the DeepLab Family of Architectures in Segmentation Internal Brachiaria Seed Structures by X-ray ImagesSeed vigor is an essential factor for the uniform emergence of seedlings and for the proper establishment of crops, being directly associated with the productive potential of agricultural crops. Accurate evaluation of this vigor is therefore fundamental for decision-making in seed management and production. Among the methods used for the analysis of physiological quality, the use of X-ray images stands out for allowing the non-destructive visualization of internal morphological structures of ... L.K. Gomes Maia, E. Freitas, W.V. Dias, J.F. Da Silva , B.D. Silva , H.F. Abud, D. G. Gomes, P. Dos Santos E Silva |
59. A Machine Learning Framework for Automated Anomaly Detection in Precision Agriculture Geospatial DataModern precision agriculture relies on the analysis of geospatial data generated by a wide range of equipment and sensors. While these datasets are foundational to data-driven management practices, they are often affected by inaccuracies arising from various sources. Existing filter systems, such as Yield Editor and Map Filter, that implement operational (e.g., abrupt changes in speed, speed limits, and removal of maneuvers), global statistical (e.g., observations that are inconsistent with t... Z.C. Komarnisky, F. Hoffmann Silva Karp |
60. Satellite Embedding-Based Corn Yield Prediction Using AutoML and Explainable AIAccurate, spatially explicit yield mapping underpins many precision agriculture decisions (e.g., variable-rate inputs and zone management), yet reliable yield monitor data are not always available and can be difficult to standardize across operations. Satellite-based yield models are often built from hand-crafted vegetation indices or phenology metrics, which may limit transferability across fields and years. Here, we evaluated a pixel-level corn yield prediction workflow that uses Satellite ... V.S. Silva, E.S. Silva, D.O. Silva, M.F. Oliveira, A.C. Tavares, R.P. Negrini, L.A. Mendes |
61. YOLOv10x-based deep learning for automated detection and counting of seeds per soybean podAccurate quantification of the number of seeds per soybean pod is a fundamental step for reliable yield estimation. However, this measurement still relies on manual procedures, which are subject to observational variability and limited scalability. In the context of digital agriculture, deep learning–based techniques have shown promise for automating the detection and counting of reproductive structures. Nevertheless, there is still limited application of models specifically aim... |
62. Row-unit Integrated Multi-camera Edge AI System for Real-time Small-grain Seeding Performance Data CollectionHigh-quality synchronized imagery collected under field conditions is a limiting factor in the development of computer vision models for small-grain seeding applications. This study presents the design, implementation, and field deployment of a row-unit integrated multi-camera data-acquisition system intended to standardize multi-view data collection during planting. The system mounts directly to a seed-drill row unit and integrates three Power-over-Ethernet (PoE) Basler cameras positioned to... A. Sharda, B. Vail, S. Rai, R. Harsha Chepally |