Conference Agenda

Near-infrared (NIR) spectroscopy (900-1700 nm) was used to develop calibration models for rapid prediction of Brix level, moisture content (MC), and computed fiber content (FC) in harvested sugarcane. Cane stalks from ten Philippine varieties harvested 10, 11, and 12 months after planting were sampled from top, middle, and bottom portions of the plant. Samples for the training set were harvested in 2019, while an independent prediction set was harvested in 2020. Cross-sectional scanning method (CSSM) and skin scanning method (SSM) were used to acquire NIR spectral data. Linear regression (LR) models performed better than partial least squares regression (PLSR) models based on values of the coefficient of determination (R²) and normalized root mean square error (NRMSE). In general, LR models for predicting Brix and MC based on spectra acquired using SSM were classified as good (NRMSE < 10%), while models for FC were considered as subpar (NRMSE approaching 20%). Prediction of ^oBrix level and MC of 2020 sugarcane samples using 2019 calibration models showed a linearity (based on R²) between predicted and measured values of 0.84 and 0.72, respectively. Regression coefficients showed clear peaks and valleys in overtone bands associated with carbonyl and hydroxyl bonds found in soluble solids and water, respectively.

Crop productivity is significantly reduced by abiotic and biotic stresses. Understanding plant stress responses requires monitoring key physiological signals, but existing methods have major limitations. They are often destructive, relying on plant extracts, or are genetically encoded sensors that are species-specific and unsuitable for real-time tracking in non-model crops. We present a novel, non-destructive nanosensor for real-time imaging of the crucial stress signal H₂O₂ in living plants. Our sensor operates in the second near-infrared window, which significantly reduces background chlorophyll autofluorescence and increases tissue penetration depth for superior imaging quality. The probe is designed as a fluorescence "turn-on" system. It co-assembles an aggregation-induced emission fluorophore (signal reporter) with a polymetallic oxomolybdate (POM) as an H₂O₂-selective quencher. In the presence of stress-induced H₂O₂, the POMs are oxidized, which deactivates their quenching ability and triggers a bright NIR-II fluorescence signal. This activatable mechanism offers a high signal-to-background ratio for clear visualization. We demonstrated the sensor's versatility by monitoring H₂O₂ signaling in various plants, including Arabidopsis, lettuce, and pepper, using both microscopy and whole-plant imaging systems. Finally, we integrated a machine learning model with the collected data to classify plant stress responses with high accuracy, showcasing a powerful new tool for plant science.

Rapid and non-destructive evaluation of fresh tea leaf quality is essential for precision harvesting and quality control. Anthocyanin content is a key quality indicator of purple tea leaves, but its accurate prediction remains challenging due to heterogeneous information from different sensors. This study developed a multi-sensor fusion approach integrating near-infrared (NIR) spectroscopy and machine vision to predict the total anthocyanin content of fresh Zijuan tea leaves.NIR spectroscopy was used to characterize internal chemical information, while visible images were used to extract color and texture features. Partial least squares regression (PLSR) and support vector regression (SVR) were applied for modeling. Variable selection was performed to reduce spectral redundancy, and feature-level fusion was compared with a model-level adaptive weighted fusion strategy driven by cross-validation error.Results showed that both NIR spectroscopy and image features could predict anthocyanin content effectively. The adaptive weighted fusion strategy demonstrated improved robustness compared with feature-level fusion. The SVR-based adaptive fusion model achieved the best overall performance, with Rp above 0.94 and RPD greater than 3.0.These findings demonstrate that adaptive model-level fusion provides a stable and practical solution for multi-sensor quality assessment of fresh tea leaves.

High‑resolution geospatial data from agricultural machinery increasingly used for real‑time control, operational documentation, and agronomic decision support. However, data usability—central to spatial data quality—depends not only on thematic, positional, temporal, and logical elements but also on whether measurements capture independent spatial information. We address a gap in prevailing standards by explicitly integrating spatial autocorrelation into quality assessment and sensor sizing for mobile operations. Using a multi‑year dataset from various field operations, we evaluate point‑level quality through ISO‑aligned elements and quantify anisotropic dependence with empirical autocorrelation, 2D variograms, Allan variance (for GNSS and actuator drift), and step‑response analysis. We then estimate the effective count of independent observations to aggregate point quality into task‑level fitness and to derive practical sizing rules for cross‑track sensor placement. Results show that effective information content is often an order of magnitude smaller than raw sample counts; thus, increasing sampling density or channel count without shortening correlation lengths rarely improves task‑level quality. Conversely, right‑sized layouts aligned with correlation structure deliver equally reliable control and documentation at lower cost and complexity. We conclude that spatial autocorrelation is a first‑class determinant of geospatial quality and should be explicitly encoded in data standards and product specifications.

Grouping plots with similar soil fertility profiles can highlight site-specific management opportunities within individual farms, especially in small-scale agriculture where managing entire plots is more feasible than subdividing them into management zones. This study evaluated whether cluster analysis can characterize soil fertility patterns by identifying groups of plots with similar multivariate attribute arrangements within farms across a coffee-growing region. Partitioning Around Medoids clustering was applied to soil fertility data (pH, OM, P, K, Ca, Mg) from 737 plots across 77 georeferenced coffee farms. The optimal number of clusters was determined using the silhouette method, where values < 0.25 indicated homogeneity. Dominant clustering drivers were identified, and regional spatial relationships among farms were examined using k-nearest neighbor networks. Soil fertility clusters per farm ranged from 1 to 5, with 60% exhibiting at least two distinct groups (k = 2). The primary clustering driver was pH, followed by Mg and Ca. Although spatial network analysis identified farm communities, no significant differences in the internal structure of plot partitioning were observed among them. Results demonstrate that it is feasible to identify groups of plots according to their soil fertility similarities, which may support farm-scale site-specific management despite the absence of clear regional patterns.

Atlantic Canada contributes approximately 73% of Canada’s total farmed shellfish production, with oysters representing a major component of the industry. Oysters require up to four years to reach marketable size, during which growth, depuration, and final quality are assessed using manual visual, tactile, and destructive testing methods. These approaches are time‑consuming, subjective, and can result in product loss. Ensuring consistent quality and food safety remains a significant challenge for the sector.

This exploratory project investigates the feasibility of using a soft X‑ray imaging system combined with machine learning techniques to non‑destructively evaluate oyster quality. Soft X‑ray imaging will be used to monitor oyster growth within the shell. Image processing and data analysis techniques will be applied to extract quality‑related parameters from the acquired images.

The study aims to assess whether soft X‑ray imaging can serve as an objective, accurate, and efficient alternative to traditional manual methods. In addition, the generated images can support food traceability and early quality screening before confirmation by conventional analytical techniques. The outcomes of this project will provide insights into the potential of soft X‑ray technology to automate oyster quality assessment, enhance economic sustainability, and strengthen food safety assurance in the shellfish industry.