Research in Agricultural Engineering - In Press
Laboratory Evaluation and Response Characterization of an Automated Soil Moisture Sensing and Control System under Varied Soil Matrices and Sensor PlacementOriginal Paper
Chailai Sasen, Anuwat Saenpong, Aphichon Mungchu, Sawanee Jansawang
This study investigated the effects of soil type, sensor depth, and sensor-to-plant distance on the performance of a solar-powered smart irrigation system. Experiments were conducted using loam, clay, and sand soils, sensor depths of 5 and 10 cm, and sensor-to-plant distances of 10 and 20 cm. Soil moisture was classified into six ranges, and a control band of 31–45% was selected to improve system stability. The results showed that the timing and volume of water dispensed by the system were strongly influenced by both soil properties and sensor placement. Increasing sensor depth from 5 cm to 10 cm increased irrigation time by approximately 29% and water consumption by nearly 68%, indicating that depth was the most influential factor. Loam soil exhibited the lowest irrigation time and water consumption, whereas clay soil required the longest irrigation duration. Sensor distance had a comparatively smaller effect, increasing irrigation demand by approximately 11%. Low standard deviation values (±(0.58–1.53) s and ±(0.06 – 0.20) x10-3 m3) confirmed good repeatability and stable system performance. The findings demonstrate that incorporating soil-dependent characteristics and appropriate sensor placement improves the consistency of automated sensor-actuator responses under laboratory conditions and provides a foundation for future field validation.
Land suitability assessment for sugarcane cultivation in Merauke regency, South Papua, Indonesia: a GIS-based maximum limitation approachOriginal Paper
Sulthan Rafi Abyan Hidayat, Mutiara Saraswati, Isna Arofatun Nikmah, Rizky Dwi Satrio
Merauke Regency in South Papua has been designated as a strategic food estate to support sugarcane (Saccharum officinarum L.) development under Presidential Decree No. 15 of 2024. Despite this mandate, no spatially explicit, multi-constraint suitability assessment integrating biophysical and legal land-use restrictions exists for this region, a critical gap given the dominance of flood-prone lowlands and extensive legally protected areas. This study addresses this gap by applying a GIS-based Maximum Limitation method, integrating agro-climatic (WorldClim), topographic (SRTM), soil (SoilGrids), and hydrological (InaRISK) data with national spatial planning (RTRW) constraints. The principal finding is that hydrological limitations dominate: 55.75% of the regency is not suitable due to flood susceptibility and poor drainage. After applying legal constraints, only 744.72 km² (1.57% of the regency) is realistically available for development, comprising 168.23 km² of highly suitable (S1) and 576.50 km² of moderately suitable (S2) land, concentrated in the southern coastal districts. These results demonstrate that presidential-scale agricultural mandates require high-resolution, multi-constraint spatial analysis to avoid ecologically and legally untenable expansion, and provide a replicable geospatial engineering framework for guiding mechanised crop development in tropical lowland food estate regions.
Integrating Sentinel-1 SAR and Sentinel-3 SLSTR Data with Random Forest Algorithm for Evapotranspiration Estimation in Malaysian Durian OrchardsOriginal Paper
Abhikrishnan Gopakumar, Siva K Balasundram, Christopher Teh Boon Sung, Wan Fazilah Fazlil Ilahi
A remote sensing based methodology is adopted to estimate the evapotranspiration in Malaysian durian orchards by integrating Sentinel-1 and Sentinel-3 data sets. Reference evapotranspiration data were obtained from the MODIS satellite and the weather station in the study area. The collected data were incorporated into a random forest machine learning algorithm to estimate the evapotranspiration. Two models were developed using the reference data collected from satellite and weather station and both models were statistically significant. The model with MODIS reference data achieved an r-value of 0.971 and an RMSE of 0.191 mm for calibration, and 0.913 and 0.885 mm for validation, respectively. The validation is performed using a temporally independent dataset drawn from three months before and after the primary sampling period. Similarly, the model with weather station reference data achieved an r-value of 0.990 and an RMSE of 0.710 mm for calibration, and 0.998 and 0.215 mm for validation, respectively. This study demonstrates the utility of high-resolution satellite data combined with artificial intelligence in estimating evapotranspiration with reasonable accuracy at a low cost.
From Rust to Resilience: A Low-Cost Computer Vision Framework for Precision Dissolved Oxygen Monitoring in Rural FarmingOriginal Paper
Adineco I. Suriaga, Paul M. Cabacungan, Archilyn S. Semanero, Alvin Joseph Macapagal, Catherine G. Lafuente, Maria Theresa Joy G. Rocamora, Bryan Kristofer A. Manabat, Nerissa G. Cabacungan, Lea Cristina D. Macaraig, Emma Porio, Carlos M. Oppus, Nathaniel Joseph C. Libatique, Gregory L. Tangonan, Arlen Sean G. Baita, Aljen Rioege P. Carmona, Patronilo A. Guzman, Nheka Louise D. De Mesa
Dissolved oxygen (DO) depletion is a primary driver of catastrophic aquaculture failure. However, commercial probes costing $1,350 (Php67,500)—representing up to 15 months of median income for smallholders—remain economically inaccessible, leaving farmers unable to monitor water quality effectively. This study introduces "Frugal Digital Volumetry," a disruptive sensing paradigm repurposing upcycled metal waste and smartphone-integrated digital microscopy into a precision instrument. The framework utilizes iron oxidation colorimetry as a biochemical proxy for DO, employing a computer vision pipeline involving RGB-to-CIE Lab* transformations and CIEDE2000 formulas to quantify oxidation levels via image analysis. Technical validation against industry sensors demonstrated high-fidelity correlation in aquaculture environments, with mean deviations between 1.88% and 2.51%. Although accuracy decreases in complex septic matrices due to turbidity, the system reliably detects critical hypoxia thresholds (3–6 mg/L) The novelty of this study lies in the validated integration of ferrous corrosion colorimetry, CIEDE2000 perceptual color difference quantification, and Frugal Innovation principles into a low-cost dissolved oxygen monitoring system deployable by non-specialist community users. Although these components have been independently investigated in prior literature, their convergence into a field-operable aquaculture monitoring instrument below the $80 cost threshold has not been previously reported. Field implementation in Pampanga, Philippines, validated the model’s socio-technical scalability, reducing capital expenditure by over 98%. This research demonstrates that AI-augmented frugal innovation can eliminate financial barriers to precision aquaculture. By providing a low-cost, high-accuracy alternative, the system enables resource-constrained communities to transition from reactive crisis management to proactive ecological resilience and sustainable food security.
Multi-Sensor VOC Monitoring for Quantifying Degassing Dynamics in Roasted Coffee BeansOriginal Paper
Sony Wardoyo, Abraham Abraham, Herbert Innah, Eva Papilaya, Yafeth Wetipo
Understanding the degassing behavior of roasted coffee beans is essential for quality evaluation and post-roasting process control. Conventional roast-level assessment largely relies on human sensory observation, which is inherently subjective and difficult to implement for real-time quality monitoring. Moreover, quantitative descriptions of volatile organic compounds (VOCs) and carbon dioxide (CO₂) emission dynamics across roasting levels remain limited. This study aims to develop a multi-sensor VOC monitoring approach to quantitatively analyze degassing dynamics in roasted coffee beans and to evaluate its capability for roast-level differentiation. Gas emissions from roasted coffee beans were measured using a combination of CCS811 and TGS822 gas sensors to capture variations in VOC concentration associated with thermal treatment during roasting. The recorded sensor signals were processed through characteristic parameter extraction to identify volatility patterns corresponding to light, medium, and dark roast levels. The experimental results demonstrate that the combined sensor responses provide clear differentiation among roasting levels and reveal a strong relationship between roasting intensity and the detected volatile emissions. The proposed multi-sensor system provides a rapid, non-destructive method for quantifying degassing behavior in roasted coffee beans and shows potential for objective roast-level classification and improved quality control in the coffee processing industry.
Investigation of Bulk Cargo Capture Process by Auger Screw Spiral Flights in the Loading Zone of Screw FeederOriginal Paper
Vasyl Vasylkiv, Dmytro Radyk, Larysa Danylchenko
This study investigates bulk-material capture within the loading zone of a screw feeder and evaluates the influence of screw geometry on loading uniformity. A specialized experimental rig with a segmented loading nozzle was developed to localize material capture along the loading zone. Cylindrical and tapered screw spirals with different inter-turn volume distributions were compared under identical operating conditions. Conventional cylindrical spirals exhibited pronounced spatial non-uniformity, with the initial screw flights capturing most of the bulk material, causing localized overloading, increased drive torque, and unstable operation. Measurements across the loading nozzle width also revealed asymmetric material intake depending on the direction of screw rotation. In contrast, tapered screw spirals with a progressively increasing inter-turn volume redistributed material more uniformly. A loading uniformity criterion (γ) was introduced to quantify the spatial distribution of material capture. Under the investigated conditions, a cone apex angle of 35–40° produced the highest loading uniformity, corresponding to the minimum γ value and nearly uniform bulk-material drawdown. The proposed methodology provides new insight into localized bulk-material capture and offers a practical basis for optimizing screw feeder geometry to improve loading uniformity, reduce drive torque, and enhance the operational stability of industrial screw feeding systems.
Effect of probe diameter, cultivar and postharvest storage on cocoa pod firmness and deformation energyOriginal Paper
Amuaku Randy, Francis Kumi, Godwin K. Amanor, Enoch Asante, Gladys Pepertual Awudi
Mechanical characterization of cocoa pods is essential for improving postharvest handling, mechanized processing, and objective quality assessment. This study investigated the effects of probe diameter, cultivar, and postharvest storage duration on cocoa pod firmness, deformation energy, and yield stress using penetration testing. Three cultivars (Amazonia, Forastero, and Amelonado) were evaluated under controlled conditions (27 ± 2°C; 70 ± 5% RH) over 8 days using probe diameters of 3.5, 8, and 11 mm. Results showed that probe diameter significantly influenced measured firmness, with smaller probes producing higher values due to stress concentration effects. Firmness and deformation energy decreased significantly during storage, particularly within the first 2-4 days, indicating rapid structural degradation. Cultivar-specific responses were observed, with Forastero exhibiting the highest mechanical strength and slowest softening, while Amazonia showed the fastest degradation, losing approximately 60% of firmness by Day 6. A strong positive correlation between firmness and deformation energy (r = 0.9478-0.9999; p < 0.05) confirms deformation energy as a reliable substitute for mechanical integrity. Partial Least Squares Regression models improved with increasing probe diameter, with optimal performance observed in Forastero (R² = 0.8118; RMSE = 1.4036 MPa).
Chacracteristics of tiger nut milk pretreated with magnetic fieldShort Communication
M.M Odewole, T.O Abodunrin, G.O Ajayi, S.O Yusuf
Thermal pretreatment of tiger nut milk (TNM) via pasteurization may have adverse effect on its quality characteristics. The use of magnetic field (MF) to pretreat TNM is a promising non-thermal method. This study investigated the effects of MF pretreatment on three (3) selected quality characteristics of TNM. Fresh TNM samples were pretreated at 0.25 - 15.37 mT MF strength under static and pulse MFs for 10 min. All samples were analyzed for calcium, vitamin E and microbial load, using standard procedures. Results showed that, MF pretreatment led to an average 10% increment in the calcium of TNM; vitamin E reduced by 2.28%, but the reduction was not significantly different from that of pasteurized TNM. Partial reduction in microbial load of TNM was achieved with MF pretreatment. MF is a viable quality-enhancing pretreatment alternative for TNM. Studies should be done on more quality characteristics of MF pretreated TNM.
Operational parameters of a CI engine running on a diesel-HVO-TME fuel blendOriginal Paper
Jakub Čedík, Radek Pražan, Petr Jevič, Jaroslav Mrázek
This paper presents an alternative fuel blend for compression ignition engines consisting of 73.2% diesel, 20% hydrotreated vegetable oil and 6.8% tallow methyl ester, with 100% diesel used as a reference. A John Deere 6620 AutoQuad tractor was used for the measurements, while the monitored parameters included performance, fuel consumption and gaseous emissions. The measurement was performed under stable engine conditions. The results indicate that when using the tested fuel blend, all the monitored operational parameters of the engine were similar to those obtained with mineral diesel, especially performance, brake thermal efficiency and CO2 emissions. There was also a slight decrease in all monitored gaseous emissions using the test fuel in comparison with diesel.
Managing Data Heterogeneity in Precision Agriculture Using Graph Attention Neural NetworksOriginal Paper
Muhammad Bello Kusharki, Muhammad Muktar Liman, Bilkisu Muhammad-Bello, Nachamada Blamah
The increasing complexity of agricultural data, stemming from diverse environmental conditions, poses a significant challenge in precision agriculture. Traditional machine learning models often struggle with data heterogeneity, leading to reduced model accuracy and limited generalizability in real-world applications. To overcome these challenges, this study leverages Graph Attention Neural Networks (GATs) to effectively model complex relationships within heterogeneous datasets. GATs utilize attention mechanisms to selectively emphasize critical features while minimizing noise, improving predictive accuracy. This research applies GATs to classify crop diseases affecting maize, rice, and wheat, using PlantVillage and PlantPAD datasets, which contain over 114,000 labeled plant disease images. Additionally, synthetic data augmentation techniques, including Generative Adversarial Networks (GANs) and domain adaptation frameworks, were employed to enhance dataset diversity and robustness. Experimental results demonstrate that GATs outperform traditional models, achieving 96.99% accuracy, compared to 87% for CNNs and 78% for SVMs. The interpretability of attention weights also provides valuable insights into the key factors influencing disease classification. Future research should focus on enhancing the scalability and computational efficiency of GATs for real-time applications in resource-limited agricultural settings. Additionally, integrating real-time sensor data streams could further improve adaptability and predictive performance, ensuring more sustainable and intelligent agricultural systems.
Automated analysis of lettuce seed primary root protrusion using computer visionOriginal Paper
Heiber Andres Trujillo, Rafael Mateus Alves, Robson Campos de Lima, Francisco Guilhien Gomes-Junior
Despite high germination rates in lettuce seeds, seedling emergence often varies in terms of speed and percentage, highlighting the need for vigor assessment during the early stages of seedling development. This study employed transillumination imaging and Machine Learning to determine the optimal time intervals for assessing seed vigor in the Roxa and Vanda genotypes. Morphological parameters such as Area, Perimeter, Circularity, and Solidity were extracted from the images using computer vision and analysed through multiple linear regressions applying the ordinary least squares (OLS) model. The results demonstrated that the proposed model effectively identified morphological traits associated with seed vigor. By identifying Circularity and Solidity as reliable early indicators of physiological performance, the study established 21 hours for the Roxa genotype and 16 hours for the Vanda genotype as the most suitable evaluation periods, showing the strongest agreement with conventional vigor assessment methods. In particular, Circularity and Solidity exhibited a strong association with seed vigor and primary root protrusion dynamics. These findings highlight the potential of automated image processing and Machine Learning as rapid, objective and non-destructive tools for seed vigor assessment, contributing to significant advances in seed technology and quality assessment systems.