Conference Agenda

A novel continuous-flow polymerase chain reaction (PCR) microfluidic chip is developed for rapid, portable, and energy-efficient DNA amplification, targeting on-site agri-food safety monitoring and field-deployable biological diagnostics. The device integrates two externally controlled heating zones with a centrally located, flow-rate–regulated cooling zone, enabling PCR thermocycling through forced convection rather than conventional bulk heating. By incorporating a water-cooling channel beneath the glass chip, a stable low-temperature annealing zone (<313 K) is achieved, allowing denaturation, annealing, and extension processes to occur sequentially within a compact continuous-flow architecture. To enhance temperature uniformity and robustness for field operation, a poly (methyl methacrylate) (PMMA) cooling channel is combined with a thin aluminum cover. The chip has a compact footprint of 76 mm × 26 mm × 3 mm, supporting portability and integration into point-of-need diagnostic platforms. Computational fluid dynamics simulations were conducted to optimize heater spacing and investigate the effects of chip materials, cooling channel operating parameters, and geometric configurations on surface temperature distribution. Experimental validation demonstrated successful amplification of 372 bp and 478 bp DNA fragments, confirming that the proposed water-cooling-assisted thermocycling strategy supports efficient DNA amplification under continuous-flow conditions.

Blackheart internal defect is a major pre-harvest disease affecting global pomegranate production. Distinguishing defective fruit externally is challenging, typically requiring expert assessment. This study aimed to develop a non-destructive classification method for blackheart detection. We evaluated 903 pomegranates (543 defective, 360 sound) using visible-near-infrared (Vis-NIR, 400–1000 nm) hyperspectral reflectance imaging and conventional measurements of physical attributes (weight, volume, density). Image processing extracted morphological (circularity, sphericity, projected surface, perimeter), colorimetric (RGB, Lab*), and Grey Level Co-occurrence Matrix (GLCM) textural parameters. A baseline Partial Least Squares Discriminant Analysis (PLS-DA) model utilising only conventional and image-derived attributes achieved classification accuracies of 80% (calibration) and 85% (prediction). To enhance classification performance, spectral reflectance intensities at five key wavelengths (450, 550, 570, 610, and 675 nm) were integrated into the model. The inclusion of these targeted spectral variables significantly improved the PLS-DA model, yielding 88% accuracy for both calibration and prediction datasets. These results demonstrate that combining optical spectral signatures with physical and image-derived morphological and textural features provides a rapid, reliable, and non-invasive approach for detecting blackheart in pomegranates. Thereby improving postharvest quality control.

Detecting foreign materials in cotton presents a significant technical challenge to improving cotton grade, particularly when transparent or semi-transparent mulching films are intermixed with cotton at different bulk densities. Studying optical properties is an effective approach to this problem. In this study, the optical properties of cotton at different bulk densities and four typical foreign materials, including hulls, cotton stems, green leaves, and mulching films, were quantitatively characterized over 400–2400 nm. A double integrating sphere-based spectroscopic measurement system combined with the inverse adding-doubling (IAD) algorithm was used to obtain the absorption coefficient (μ_a), reduced scattering coefficient (μ_s’), and scattering anisotropy factor (g). System accuracy was validated using liquid phantoms, with mean relative errors of 21.00% for μ_a and 10.86% for μ_s’. Significant differences in optical properties were observed between cotton and foreign materials. Cotton bulk density was positively correlated with μ_a and μ_s’, but negatively correlated with g. Random Frog (RF) and Variable Combination Population Analysis (VCPA) were used to select feature bands. Random forest classification based on the combined optical properties reached 100% for both full wavelengths and feature bands. These findings provide a theoretical foundation and data support for online optical sensing of foreign materials in cotton processing.

Micronaire is a key indicator of cotton fiber quality. In cotton procurement, airflow-based micronaire measurement is affected by operator-related variability, and deviations in sample mass from the nominal standard can cause systematic drift in differential-pressure signals, thereby reducing measurement accuracy. To address this problem, a cotton fiber micronaire measurement method combining mass-pressure differential compensation and machine learning was developed on an airflow testing platform. Based on differential-pressure and sample-mass data, a coupled compensation model was established, achieving R² = 0.99, RMSE = 31.28 Pa, and MAE = 23.90 Pa for 492 samples. Using compensated pressure difference and sample mass as input features, multiple regression models, including linear regression, polynomial regression, Ridge regression, support vector regression (SVR), K-nearest neighbors, and random forest, were developed and compared. Results showed that SVR was the best-performing model, with R² = 0.96, RMSE = 0.05, and MAE = 0.04 on a 101-sample test set. The proposed method improves the robustness and accuracy of micronaire measurement under the nominal 10 g standard and provides support for online cotton quality inspection during procurement and processing.

Effectiveness of plant disease management relies in the early detection of the numerous plant pests regularly affecting the crops. However, the timely identification requires frequent monitoring, specialized laboratory equipment, and qualified personnel, making it a time demanding and costly process. That’s why diseases are often detected late, making their control more difficult and expensive.

In study we introduce ADDAMS, a command-line tool designed for efficient detection of plant diseases in agricultural crops. ADDAMS requires digital images (from multispectral or phone cameras) as the only input and executes a pixel-level analysis by comparing images to disease-specific spectral profiles, obtained by measuring reflectance of diseased plants through a portable spectroradiometer (FieldSpec® 3).

ADDAMS enabled the identification of viral infections, fungal diseases, and mite infestations in grapevine, artichoke and melon in open field conditions. The highest accuracy was achieved in the presence of a single disease, but excellent results were also obtained when multiple diseases occurred simultaneously. This is a crucial aspect and allows ADDAMS to be applied to crops affected by multiple pests.

ADDAMS is therefore a strong candidate for use as a decision-support tool in field diagnostics, capable of anticipating diseases detections and helping reduction of plant protection products applications.

South Korea has the second highest per capita garlic consumption in the world, leading to the development of various harvesters for garlic cultivation.. However, a supplementary operation is required to remove foreign matter such as soil and stones generated during the harvesting process. To overcome this, the development of a foreign object sorting device for garlic harvesters was pro-posed. Accordingly, this study conducted fundamental experiments to optimize the foreign object sorting device. First, a pneumatic cylinder-driven impact-type sorting test device was constructed for experimentation. This test device was con-nected to the rear of two types of garlic harvesters for testing. An ultra-high-speed camera was used to capture the movement trajectories of foreign object for each factor (type, size, and conveying speed). These trajectories were plotted to deter-mine the optimal position for the sorting device and to analyze its foreign object removal effectiveness. Additionally, the analysis determined whether the impact device's operating speed was appropriate when large quantities of foreign object were introduced. Based on these findings, the sorting device will be improved. Future plans include developing an automated foreign object sorting system by integrating it with an image recognition-based foreign object detection device.

Unlike broilers, AI technologies are available to a limited extend for laying hens due to application constraints on real-farm conditions. In recent decades, animal welfare assessment has evolved from general herd level management toward a holistic framework prioritizing the mental states of individual animals. However, monitoring the integument state manually is highly time-consuming and subjective. As a result of these facts, the objective of this study was to develop a sensor prototype driven AI models to monitor integument damage of individual laying hens. The experiments were conducted on a typical organic laying hen farm in Mecklenburg-Western Pomerenia, Germany: A mobile camera sensor prototype was placed in the litter area and used primarily for grooming and resting that allowed continuous, stress-free data collection. Additionally, regular manual assessment of integument damages on dorsal and ventral body areas identified suitable target markers and target positions for the camera model. Using the collected data, high-quality training sets were created by combining deep learning and large language models (LLM) within a Python-based interface. Manual correction created high-quality training data sets for specialized AI models. As a future work, it is aimed to transform the prototype into a smart early warning tool in regards to animal welfare.