Conference Agenda

Innovations in canopy-specific defoliation can improve cotton (Gossypium hirsutum L.) harvest efficiency and fiber quality across diverse production environments. This study evaluated the effects of sequential canopy defolia-tion on cotton fiber properties using multivariate analysis at two contrasting field sites in South Carolina, USA. Field trials were conducted in 2024 at the Edisto Research and Education Center and the South Carolina State Univer-sity Research and Demonstration Farm using cultivar DP 2127 B3XF in a randomized complete block design. Treatments included five sequential de-foliation intervals (15, 10, 8, 5, and 3 days between canopy layers) and broadcast control. Defoliants were applied using a robotic ground sprayer with pulse-width-modulated nozzles for canopy-targeted delivery.

Fiber quality was assessed using High Volume Instrument and Advanced Fi-ber Information System measurements. Principal component analysis showed that the first two components explained over 90% of total variance across both sites. Shorter defoliation intervals (3–8 days) improved fiber brightness, uniformity, and cleanliness, while longer intervals (10–15 days) increased trash content and nep formation. Results indicate that defoliation timing had a greater influence on fiber quality than site conditions, demon-strating the value of sequential canopy defoliation combined with precision mechanized applications.

Assessment of the levels of agricultural mechanization of rice producing regions (1,5,7,8.11 and 12) in the Philippines using the Modified Agricultural Mechanization Index (MAMI) was conducted. Results showed that rice production in the country are still male-dominated (71% male, 29% female) with men performing most labor-intensive tasks Most of the operations utilize three primary power sources: human, man-animal power, and man-machine system. Land preparation and harvesting operations are predominantly done by man-machine system while crop establishment and crop care operations are still done manually.

Results further showed that the level of mechanization for the six regions were calculated at 4.099 hp/ha for Region 1, 1.896 hp/ha for Region 5, 2.509 hp/ha for Region 7, 1.287 hp/ha for Region 8, 4.463 hp/ha for Region 11 and for Region 12, respectively, Values are still way below the ideal level of mechanization. Machinery gaps for each farm operation were also calculated based on the actual level to attain the ideal level of mechanization. These findings could guide decision makers and planners advocating agricultural mechanization to propel rice production in the country.

As the growth of sensors and software becomes operational, agriculture’s management and decision-making grapple with the selection of technologies from a broader range of options to increase yield, efficiency, and profitability. Generally, technology adoption for agricultural applications pursues a more efficient use of inputs, while production costs are constrained and yields maximized. This scenario can be empirically driven by the decision-makers’ beliefs rather than an effective benefit-cost evaluation. This study evaluated the production cost of 18 grain production systems. Simulations of higher-efficiency on fuel, pesticides, labor, irrigation, fertilizer, seed, field efficiency, and components of administrative costs were made to support forecasts on cost reduction. The estimations made considered the viability of technologies aiming to increase yield, such as biotechnology adoption or slow-release fertilizers. The impact of inputs’ costs and the grain market price was also addressed by considering distinct producing periods. Irrigated scenarios were able to pay more for technological adoption that improve efficiency of all studied variables, except for herbicide spraying. Also, high-yield years would be those seasons in which investing on technologies to increase production would be more prone to fail due to the lack of economic incentives such as low grain prices.

Moringa oleifera is a fast-growing multi-functional crop increasingly cultivated in Mediterranean environments for biomass production. Reliable above-ground biomass (AGB) estimation is essential for precision management. However, conventional plot-based spectral extraction often introduces soil and background interference, limiting model robustness. This study proposes a crown-level feature extraction pipeline for AGB modelling based on individual canopy delineation. The research was conducted in an experimental field in Sicily, characterized by four genotypes and monitored across three growth stages. High-resolution RGB and multispectral imagery were acquired using an unmanned aerial vehicle (UAV). Photogrammetric processing generated a dense 3D point cloud, enabling the derivation of a digital elevation model (DEM) and a canopy height model (CHM) through surface normalization. Vegetation and non-vegetation classes were separated at the point-cloud level, and an automatic crown delineation algorithm was applied to the CHM to identify individual tree objects. Spectral indices and structural metrics were then extracted at canopy scale to support biomass modelling. The segmentation-driven approach reduced background-induced variance and improved the consistency of structural and spectral predictors compared to conventional pixel aggregation. The proposed framework enhances model robustness and establishes a scalable pipeline for predictive modelling and large-scale wall-to-wall AGB mapping in high-density perennial crops.

Agricultural machinery generates complex multivariate time-series data through Controller Area Network (CAN) bus systems, comprising dozens of heterogeneous signals with varying sampling rates, temporal gaps, and strong inter-signal dependencies. Detecting anomalies in such data is critical for predictive maintenance yet challenging due to high dimensionality, irregular coverage across signals, and the scarcity of labeled fault data.

We propose a semi-supervised anomaly detection framework based on the iTransformer architecture. The model is trained exclusively on labeled nominal operation data, requiring no anomaly examples. Unlike conventional transformers that apply attention along the time axis, iTransformer treats each signal's temporal history as a single token and applies attention across signals, explicitly capturing cross-channel correlations among all 39 CAN signals simultaneously. Anomaly scores are derived from multi-step prediction errors, flagging windows where observed behavior deviates from learned normal dynamics.

On a real-world agricultural dataset, our model achieves R² scores above 0.9 on the validation set, demonstrating that iTransformer successfully learns the interdependencies governing normal machine operation. These results indicate that multivariate forecasting quality is sufficient to serve as a reliable basis for anomaly detection, offering a practical approach to fault identification that requires only nominal-class labels.