Efficient management of agricultural water resources has become increasingly critical due to climate variability and rising global food demand. This paper presents a comprehensive IoT-based system for real-time agricultural water forecasting, integrating field-deployed sensors, cloud infrastructure, and advanced machine learning models. The system automates data collection, preprocessing, and model training, enabling accurate and scalable irrigation management. We evaluate three models: a lightweight XGBoost regressor for edge deployment, a Long Short-Term Memory (LSTM) network for capturing temporal patterns, and a hybrid LSTM–XGBoost model that combines the strengths of both. The hybrid model achieved the best performance with a Root Mean Squared Error (RMSE) of 0.01705 and a coefficient of determination (R2) of 0.95, outperforming the standalone XGBoost (RMSE = 0.0184, R2= 0.92) and LSTM (RMSE = 0.0704, R2= 0.86) models. Operational insights regarding system latency, data reliability, and field maintenance are also discussed, emphasizing the model’s robustness and practical deployment potential. The results underscore the viability of data-driven irrigation forecasting for improving agricultural sustainability and optimizing resource efficiency.

Understanding appetitive fish behaviour patterns represents a critical challenge for developing intelligent aquaculture monitoring and feeding systems. We present a novel robust analytical framework that combines non-invasive monitoring techniques from 3D multibeam sonar data and environmental monitoring (temperature, dissolved oxygen, and current velocity). Our methodology introduces hourly-averaged acoustic backscatter profiles to quantify behavioural metrics, including the maximum biomass depth and the biomass boundary, enabling comprehensive characterisation of vertical fish distribution dynamics. We used k-means clustering to identify distinct high- and low-activity behavioural modes. The analysis demonstrates appetitive vertical patterns, with fish aggregating deeper in the cage during non-feeding times, moving to shallower depths before scheduled feeding, staying shallow in the water column after feeding and showing high-activity behaviour through the whole cage during feeding hours. These behavioural patterns reveal new insights by correlating vertical fish distribution patterns with scheduled feeding times, suggesting measurable appetitive responses using a multibeam sonar. The robustness of those findings was evaluated through seasonal comparisons, environmental perturbation filtering, and consistency analysis across multiple months. This included storm conditions with lower surface temperatures and stronger currents than typical for the area, where fish consistently retreated to maximum depth. This integrated and explainable inference pipeline provides a foundation for future applications in feeding behaviour detection and optimisation. The work advances the development of non-invasive, data-driven tools for more efficient and sustainable aquaculture management, supporting real-time decision support systems that can optimise feeding strategies and improve fish welfare through behaviour-based management while addressing the industry’s challenge of significant feed loss.

As aquaculture systems become increasingly complex, simulation tools must evolve to support intuitive, reliable, and domain-specific exploration. In our earlier work, we developed a web-based platform that combined Functional Mock-up Units (FMUs), surrogate modeling techniques, and a basic chatbot assistant to simulate key components of aquaculture systems such as fish growth, water quality, and behavior. While this platform improved simulation efficiency and usability, it lacked the contextual awareness and flexibility needed to fully support users in configuring and understanding complex simulation scenarios. In this paper, we present an enhanced, robust agentic system that builds on the previous platform by introducing a more intelligent, domain-aware AI assistant. This assistant is powered by a knowledge-augmented generative architecture (KAG-RAG), which integrates a Neo4j-based knowledge graph constructed from aquaculture-specific tutorials and documentation with a Large Language Model (LLM). The result is an assistant capable of interpreting natural language queries, retrieving and reasoning over relevant context, and guiding users through simulation setup and analysis in real time. To improve the overall user experience, the system provides more context-aware assistance to help users configure simulations correctly and avoid common setup errors. We compare our enhanced agent with two baselines: one using only the OpenAI API and another using a standard RAG setup. Our findings show that the agent delivers more accurate, relevant, and context-aware responses. It better understands complex, multi-step queries, uses aquaculture terminology more effectively, and provides clearer, more helpful guidance during simulation. Overall, this agentic system offers a more powerful, adaptive, and user-friendly environment for researchers, policymakers, and industry professionals, supporting the development of sustainable, efficient, and economically viable aquaculture practices.

In the rapidly expanding field of aquaculture, ensuring the welfare of fish is essential for sustainability and productivity. Recognizing how environmental conditions and operational practices impact fish health, there is a need for robust methodologies that can predict and assess fish welfare effectively. This paper introduces a systematic modeling approach designed to advance fish welfare assessment by integrating diverse data sources including historical, environmental, operational data, and time-series analysis into a cohesive analytical framework. The proposed model focuses on developing and implementing inference rules that extract meaningful insights from heterogeneous datasets collected from multiple sources, such as direct observations, sensor outputs, and operational logs. This integration facilitates the use of machine learning techniques to assess and predict the welfare of fish groups. Time-series data provide a baseline for evaluating current conditions, while environmental and operational data contribute to realtime welfare assessments. In addition, the model features the construction of a fish welfare journal that significantly enhances its reasoning capabilities by providing deeper insights into the welfare status of fish over time. The utility of the model is demonstrated through its ability to predict and forecast welfare scores based on changes in environmental and operational conditions, which is vital for effective decision making, planning, and scheduling in fish farming operations. We showcase the practical applications of our model with a numerical example that illustrates its use in a real-world scenario. This example highlights how integrated data and inference methodologies can be employed to deduce welfare scores and make informed decisions about fish farming practices. Taking this example further, we train a predictive model to forecast welfare scores, illustrating the benefits of the proposed model, which not only dynamically calculates welfare scores, but also integrates time-series historical data to assess the performance of fish farming operations. Overall, this work not only deepens our understanding of fish welfare dynamics but also helps establish a more refined and robust welfare assessment protocol, extending the predictive capabilities into future operational planning and management.

The aquaculture industry faces critical challenges in optimizing feeding strategies to enhance fish growth while minimizing environmental impacts and ensuring economic viability. Traditional feeding methods often fall short in adapting to dynamic environmental conditions and fish growth rates, leading to suboptimal growth, waste, and environmental degradation. To address these issues, this study introduces a robust machine learning-based framework designed to optimize feeding processes in aquaculture. The framework employs advanced regression models such as Gradient Boosting Regressor, Elastic Net Regression, and Support Vector Regression to predict optimal feeding rates with high accuracy and efficiency. Our methodology integrates real-time data from environmental sensors, video analytics, and manual logging to predict the optimal feed amount. The goal of this comprehensive approach is to achieve high growth performance indicators such as Specific Growth Rate (SGR), Relative Growth Index (RGI), and optimal Feed Conversion Ratio (FCR), while also ensuring minimal feed spillage. By employing machine learning, we can dynamically adjust feeding amounts based on fish appetite and environmental conditions, thus ensuring sustainable and economically viable fish farming practices. This paper details the implementation of this framework, encompassing data collection and cataloging, model training, selection, and validation processes, and discusses the significant improvements over traditional methods. Our results demonstrate the model’s effectiveness in reducing waste and enhancing fish growth, illustrating the potential for wider application within the aquaculture industry.

This paper presents a 3D multiple object detection and tracking framework for identifying and quantifying changes in fish behaviour through tracking the 3D position, distance and speed of fish with respect to an underwater stereo camera. The framework consists of six essential modules based on 3D object detection to identify fish and multiple object tracking algorithms to track the fish in sequential frames. In particular, the latest version of Yolo (Yolov7) is utilised for object detection and the deep SORT algorithm is used for multiple object tracking. The framework was tested using videos captured from an underwater stereo camera in an industrial-scale sea-based fish farm. The results showed that the framework was able to accurately detect and track multiple fish in 3D. The fish position, distance and speed relative to the camera were also successfully detected. The results of this study demonstrate the effectiveness of this framework in identifying and quantifying changes in fish behaviour. The proposed novel framework has the potential to greatly enhance our understanding of fish behaviour in their natural habitats, leading to new insights into fish ecology and behaviour, while at the same time, it can enable researchers to study fish behaviour in a more detailed and accurate way.

We propose a web-based platform of integrated intelligent agents that incorporates multiple Functional Mock-up Units (FMUs) and surrogate modelling techniques. In this platform, each FMU envelops a stand-alone simulation component that represents an aquaculture system, such as the fish growth model, water quality model, and fish behavior model. Some FMUs may be computationally expensive to simulate or have different time step intervals, making integration with other FMUs difficult. To address these challenges, we employed surrogate models to substitute the more computationally expensive models. In this work, surrogate models are trained using simulation data and selected based on robustness analysis to ensure the overall system input-output reliability. The platform also includes a Chatbot component utilizing natural language processing and decision-making techniques to interpret user requests and provide tailored FMU configurations, enhancing the user simulation experience. Overall, the proposed platform provides a comprehensive and Efficient approach to modelling and simulating complex systems such as fish farms. Robustness analysis ensures the platform's accuracy and reliability, while the user-friendly interface enables easy tailored experimentation. By providing a framework for exploring the potential of aquaculture as a key source of food and income, the proposed platform represents a valuable interactive simulation tool for researchers, policymakers, and industry professionals seeking to improve the sustainability, efficiency, and economic viability of the aquaculture industry.

A key driver to ofering smart services is an infrastructure of Cyber-Physical systems (CPS)s. By definition, CPSs are intertwined physical and computational components that integrate physical behaviour with computation. The reason is to autonomously execute a task or a set of tasks providing a service or a list of end-users services. In real-life applications, CPSs operate in dynamically changing surroundings characterized by unexpected or unpredictable situations. Such operations involve complex interactions between multiple intelligent agents in a highly non-stationary environment. For safety reasons, a CPS should withstand a certain amount of disruption and exert the operations in a stable and robust manner when performing complex tasks. Recent advances in reinforcement learning have proven suitable for enabling multi-agents to robustly adapt to their environment, yet they often depend on a massive amount of training data and experiences. In these cases, robustness analysis outlines necessary components and specifications in a framework, ensuring reliable and stable behaviour while considering the dynamicity of the environment. This paper presents a combination of multi-agent reinforcement learning with robustness analysis shaping a cyber-physical system infrastructure that reasons robustly in a dynamically changing environment. The combination strengthens the reinforcement learning, increasing the reliability and flexibility of the system by applying robustness analysis. Robustness analysis identifies vulnerability issues when the system interacts within a dynamically changing environment. Based on this identification, when incorporated into the system, robustness analysis suggests robust solutions and actions rather than optimal ones provided by reinforcement learning alone. Results from the combination show that this infrastructure can enable reliable operations with the flexibility to adapt to the changing environment dynamics.

Aquaculture is one of the largest, and fastest growing industries in Norway. Recently, the industry has experienced significant development in the daily operations acquiring new technologies and systems that capture data and automate the different processes. These emerging technologies enable the generation of enormous amounts of data from sensors in the fish cages, cameras, boats, and feeding control rooms. Additional information relevant to the aquaculture industry is based on e-mails, manual notes, or intrinsic experiences and knowledge exchanges. One of the critical aspects of successful fish farming operation management, which is yet not achieved, is to allow domain experts to gain insight into the interconnection between the broad spectrum of heterogeneous data currently realized. This paper describes a framework for storing and retrieving critical information connected to fish farming based on a graph database approach. The overall architecture is presented with detailed illustrations of how data is visualized and interpreted through a user-friendly interface. Accordingly, this work demonstrates how aquaculture users can benefit from the system to identify possible connections in the data and reveal previously undiscovered causalities and correlations that suggest optimal actions. Further, studies and evaluations of the querying system are conducted, evaluating the capability of the proposed design to process complex relationships. This work showcases that the system helps fish farmers and aquaculture users gain knowledge, reveal hidden links in the data, and improve aquaculture operations.

Image segmentation is one of the key components in systems performing computer vision recognition tasks. Various algorithms for image segmentation have been developed in the literature. Among them, more recently, deep learning algorithms have been remarkably successful in performing this task. A downside with deep neural networks for segmentation is that they require a large amount of labeled dataset for training. This prerequisite is one of the main reasons that led researchers to adopt data augmentation approaches in order to minimize manual labeling efforts while maintaining highly accurate results. This paper uses classical non-deep learning methods for background extraction to increase the size of the dataset used to train deep learning attention segmentation algorithms when images are presented as time-series to the model. The method presented adopts the Gaussian mixture-based (MOG2) foreground-background segmentation followed by dilation and erosion to create masks necessary to train the deep learning models. It is applied in the context of planktonic images captured in situ as time series. Various evaluation metrics and visual inspection are used to compare the performance of the deep learning algorithms. Experimental results show higher accuracy achieved by the deep learning algorithms for time-series image attention segmentation when the proposed data augmentation methodology is utilized to increase the training dataset.

Deep convolutional neural networks have proven effective in computer vision, especially in the task of image classification Nevertheless, the success is limited to supervised learning approaches, requiring extensive amounts of labeled training data that impose time-consuming manual efforts. Unsupervised deep learning methods were introduced to overcome this challenge. The gap, however, towards achieving comparable classification accuracy to supervised learning is still significant. This paper presents a deep learning framework for images of planktonic organisms with no ground truth or manually labeled data. This work combines feature extraction methods using state-of-the-art unsupervised training schemes with clustering algorithms to minimize the labeling effort while improving the classification process based on essential features learned by the deep learning model. The models utilized in the framework are tested over existing planktonic data sets. Empirical results show that unsupervised approaches that cluster the data based on the deep learning model’s feature space representations improve the classification task and can identify classes that have not been seen during the learning process.

Real-world autonomous systems are often controlled using conventional model-based control methods. But if accurate models of a system are not available, these methods may be unsuitable. For many safety-critical systems, such as robotic systems, a model of the system and a control strategy may be learned using data. When applying learning to safety-critical systems, guaranteeing safety during learning as well as testing/deployment is paramount. A variety of different approaches for ensuring safety exists, but the published works are cluttered and there are few reviews that compare the latest approaches. This paper reviews two promising approaches on guaranteeing safety for learning-based robust control of uncertain dynamical systems, which are based on control barrier functions and control Lyapunov functions. While control barrier functions provide an option to incorporate safety in terms of constraint satisfaction, control Lyapunov functions are used to define safety in terms of stability. This review categorises learning-based methods that use control barrier functions and control Lyapunov functions into three groups, namely reinforcement learning, online and offline supervised learning. Finally, the paper presents a discussion of the suitability of the different methods for different applications.

Autonomous cyber-physical systems, CPS, in dynamic environments must work impeccably. The cyber-physical systems must handle tasks consistently and trustworthily, i.e., with a robust behavior. Robust systems, in general, require making valid and solid decisions using one or a combination of robust reasoning strategies, algorithms, and robustness analysis. However, in dynamic environments, data can be incomplete, skewed, contradictory, and redundant impacting the reasoning. Basing decisions on these data can lead to inconsistent, irrational, and unreasonable cyber-physical systems’ movements, adversely impacting the system’s reliability and integrity. This paper presents the assessment of robust reasoning for autonomous cyber-physical systems in dynamic environments. In this work, robust reasoning is considered as 1) the capability of drawing conclusions with available data by applying classical and non-classical reasoning strategies and algorithms and 2) act and react robustly and safely in dynamic environments by employing robustness analysis to provide options on possible actions and evaluate alternative decisions. The result of the research shows that different common existing strategies, algorithms and analyses can be provided together with a comparison of their applicabilities, benefits, and drawbacks in the context of cyber-physical systems operating in dynamically changing environments. The conclusion is that robust reasoning in cyber-physical systems can handle dynamic environments. Moreover, combining these strategies and algorithms with robustness analysis can support achieving robust behavior in autonomous cyber-physical systems while operating in dynamically changing environments.

In this paper, we propose a deep learning instance segmentation framework for particle extraction of microscopic images that aims at calculating planktonic species distribution and concentration in-situ. The framework comprises three essential functional tasks on in-situ time-series images collected from an autonomous underwater vehicle: 1) manual labeling of the captured images, 2) object localization, segmentation, and identification, and 3) class distribution and planktonic organisms concentration calculation. Our proposed framework is based on the mask R-CNN architecture provided by the Detectron2 library developed by Facebook Artificial Intelligence Research (FAIR) for instance segmentation. Due to its modular design, we compare the performance of different networks by alternating the backbone sub-network in order to choose the most suitable architecture for the task of instance and semantic segmentation. We compile a custom annotated dataset from planktonic time-series images and train the different models over this dataset to perform the instance semantic segmentation. Evaluation results of the proposed framework, utilizing the best performing deep learning architecture along with the new annotated dataset, show better performance in terms of speed and accuracy of both in-situ segmentation and classification compared to traditional segmentation methods. In addition, we observe a significant improvement in the object classification quality when we train the model over our newly annotated dataset instead of training it over the dataset generated from the traditional methods. The inferred data from our novel instance segmentation framework, which provides the particle class distribution and concentration, can then be used to assist in constructing a dynamic probability density map of planktonic communities dispersion and abundance.

Abstract This paper introduces a new constraint domain for reasoning about data with uncertainty. It extends convex modeling with the notion of p-box to gain additional quantifiable information on the data whereabouts. Unlike existing approaches, the p-box envelops an unknown probability instead of approximating its representation. The p-box bounds are uniform cumulative distribution functions (cdf) in order to employ linear computations in the probabilistic domain. The reasoning by means of p-box cdf-intervals is an interval computation which is exerted on the real domain then it is projected onto the cdf domain. This operation conveys additional knowledge represented by the obtained probabilistic bounds. The empirical evaluation of our implementation shows that, with minimal overhead, the output solution set realizes a full enclosure of the data along with tighter bounds on its probabilistic distributions.