Automation of logistics enhances efficiency, reduces costs, and minimizes human error. Image processing—particularly vision-based AI—enables real-time tracking, object recognition, and intelligent decision-making, thereby improving supply chain resilience. This study addresses the challenge of deploying deep learning-based object detection on resource-constrained embedded platforms, such as NVIDIA Jetson devices on UAVs and ground robots, for real-time logistics applications. Specifically, we provide a comprehensive comparative analysis of YOLOv5 and YOLOv8, evaluating their performance in terms of inference speed, accuracy, and dataset-specific metrics using both the Common Objects in Context (COCO) dataset and a novel, custom logistics dataset tailored for aerial and ground-based logistics scenarios. A key contribution is the development of a user-friendly graphical user interface (GUI) for selective object visualization, enabling dynamic interaction and real-time filtering of detection results—significantly enhancing practical usability. Furthermore, we investigate and compare deployment strategies in both Python 3.9 and C# (ML. NET v3 and .NET Framework 7) environments, highlighting their respective impacts on performance and scalability. This research offers valuable insights and practical guidelines for optimizing real-time object detection deployment on embedded platforms in UAV- and ground robot-based logistics, with a focus on efficient resource utilization and enhanced operational effectiveness.

Some chemical reactors exhibit coupled dynamics with multiple equilibrium points and strong nonlinearities. The accurate modeling of these dynamics is crucial to optimal control and increasing the reactor’s economic performance. While neural networks can effectively handle complex nonlinearities, they sacrifice interpretability. Alternatively, block-oriented Hammerstein–Wiener models and Koopman operator-based linear predictors combine nonlinear representation with linear dynamics, offering a gray-box identification approach. This paper comprehensively reviews recent advancements in both the Hammerstein–Wiener and Koopman operator methods and benchmarks their accuracy against neural network-based approaches to modeling a large-scale industrial Fluid Catalytic Cracking fractionator. Furthermore, Monte Carlo simulations are employed to validate performance under varying signal-to-noise ratios. The results demonstrate that the Koopman bilinear model significantly outperforms the other methods in terms of accuracy and robustness.

In this paper, we aimed to identify the dynamics of a crude distillation unit (CDU) using closed-loop data with NARX−NN and the Koopman operator in both linear (KL) and bilinear (KB) forms. A comparative analysis was conducted to assess the performance of each method under different experimental conditions, such as the gain, a delay and time constant mismatch, tight constraints, nonlinearities, and poor tuning. Although NARX−NN showed good training performance with the lowest Mean Squared Error (MSE), the KB demonstrated better generalization and robustness, outperforming the other methods. The KL observed a significant decline in performance in the presence of nonlinearities in inputs, yet it remained competitive with the KB under other circumstances. The use of the bilinear form proved to be crucial, as it offered a more accurate representation of CDU dynamics, resulting in enhanced performance.