Transformable semantic map based navigation using autonomous deep learning object segmentation

Geospatial Artificial Intelligence (GeoAI) is an important component in the analysis of multi temporal satellite imagery in order to comprehend spatial changes in dynamic environments. In the present paper, Deep Geo Detect is introduced as a hybrid deep learning model that is aimed at automated geographic change detection by combining Convolutional Neural Networks (CNNs) and a Random Forest (RF) classifier. The approach suggested is a patch based learning scheme on multi spectral satellite image where the CNN element is used to learn hierarchical spatial spectral representations using bi-temporal image pairs. A Random Forest classifier then uses these deep representations to improve the stability of decisions and deal with the interactions between features. To overcome the difficulties that are normally related to change detection, such as class imbalance and spatial heterogeneity, a hierarchical patch labeling mechanism and data balancing plan are included in the training pipeline. The framework can be used to capture both localized and large scale environmental changes, thus making it scalable to various geographic settings.The proposed architecture can be used to develop a scalable and interpretable remote sensing based change analysis system by integrating deep feature learning with ensemble based classification.

Deep learning has emerged as a powerful tool in various domains, revolutionising machine learning research. However, one persistent challenge is the scarcity of labelled training data, which hampers the performance and generalisation of deep learning models. To address this limitation, researchers have developed innovative methods to overcome data scarcity and enhance deep model learning capabilities. Two prevalent techniques that have gained significant attention are transfer learning and self-supervised learning. Transfer learning leverages knowledge learned from pre-training on a large-scale dataset, such as ImageNet, and applies it to a target task with limited labelled data. This approach allows models to benefit from the learned representations and effectively transfer knowledge to new tasks, resulting in improved learning performance and generalisation. On the other hand, self-supervised learning focuses on training models using pretext tasks that do not require manual annotation, allowing them to learn valuable representations from large amounts of unlabelled data. These learned representations can then be fine-tuned for downstream tasks, mitigating the need for extensive labelled data. In recent years, transfer and self-supervised learning have found applications in various fields, including medical image processing, video recognition, and natural language processing. These approaches have demonstrated remarkable achievements, enabling breakthroughs in areas such as disease diagnosis, object recognition, and language understanding. However, while these methods offer numerous advantages, they also have limitations. For example, transfer learning may face domain mismatch issues between the pre-training and target domains, while self-supervised learning requires careful design of pretext tasks to ensure meaningful representations. This review paper explores the recent applications of these pre-training methods in various fields within the past three years. It delves into the advantages and limitations of each approach, assesses the performance of models employing these techniques, and identifies potential directions for future research. By providing a comprehensive review of current pre-training methods, this article offers guidance for selecting the best technique for specific deep learning applications to address the data scarcity issue.

Purpose The purpose of this paper is to enrich the debate on business models for sustainability in contingent situations. Approach We employ literature on business model and business model for sustainability theory and a contingency framework advanced in previous literature, and apply qualitative methods of multiple cases of four solar companies in two distinct dynamic business environments to examine how business models develop to align with changes in the business environment. Findings We provide empirical evidence that confirm and support for the contingency framework on business model dynamics and how it can be extended. 1) Adding a conceptual framework helps us understand the roles of the components in a more detail, which reveals that 2) in a hypercompetitive situation there is a focus on adapting the customer interface and business structure, while in an environmental shift situation innovating the value proposition and revenue model is needed. 3) Overall value creation activities should not be limited to the value proposition, intangible values in the customer interface also need to be addressed suggesting customer sensitivity in the customer interface. as a way to better understand the interaction between the firm and the customer in the customer interface. Originality Jointly, our findings provide new insights on business model for sustainability changes in dynamic business environments thereby the spread of and scale up of sustainable technologies.

At present, several feature meta-models have been come up with. However, they can’t meet the requirements of dynamic Internet environment or software reuse. This paper proposes a feature meta-model based on ontology as well as its formal description. Meanwhile, FTM (Flexible Transaction Model) mechanism is considered. In particular, it is adaptable to the changes in dynamic environment and can meet the requirement of software reuse. Finally, an example is given to verify this model.KeywordsFormal DescriptionFormal SemanticPropositional FormulaSoftware ReuseBusiness ActionThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Reliability assessment considering stress drift and shock damage caused by stress transition shocks in a dynamic environment

The dependency on extensive expert knowledge for defining subgoals in hierarchical reinforcement learning (HRL) restricts the training efficiency and adaptability of HRL agents in complex, dynamic environments. Inspired by human-guided causal discovery skills, we proposed a novel method, Human Causal Perception and Inference-driven Hierarchical Reinforcement Learning (HCPI-HRL), designed to infer diverse, effective subgoal structures as intrinsic rewards and incorporate critical objects from dynamic environmental states using stable causal relationships. The HCPI-HRL method is supposed to guide an agent's exploration direction and promote the reuse of learned subgoal structures across different tasks. Our designed HCPI-HRL comprises two levels: the top level operates as a meta controller, assigning subgoals discovered based on human-driven causal critical object perception and causal structure inference; the bottom level employs the Proximal Policy Optimisation (PPO) algorithm to accomplish the assigned subgoals. Experiments conducted across discrete and continuous control environments demonstrated that HCPI-HRL outperforms benchmark methods such as hierarchical and adjacency PPO in terms of training efficiency, exploration capability, and transferability. Our research extends the potential of HRL methods incorporating human-guided causal modelling to infer the effective relationships across subgoals, enhancing the agent's capability to learn efficient policies in dynamic environments with sparse reward signals.

Multi-factor fusion index assessment of digital speckle interferometry images quality under dynamic environment

Experimental investigation on the heating and cooling performance of a solar air composite heat source heat pump

Reasoning about environmental change in dynamic environments is a key factor in predicting an agent's next move in those environments. Cultural Algorithms provide a mechanism to reason about environmental dynamics. In this study, cultural algorithms show encouraging results when applied to environments where the problem is to find the highest peak in a multidimensional landscape, where the peaks are moving over time. Here, we use De Jong's environmental dynamics simulator and observe how tracking of change is affected by the frequency of change, and the magnitude for both a self-adaptive EP and a cultured EP version. It is shown that the cultured system is less sensitive to the environmental changes and outperforms the self-adaptive EP system particularly when the frequency of change becomes very large.

With the rapid advancement of robotics and artificial intelligence, embodied intelligent agents operating in dynamic and unstructured environments have emerged as a key technological direction within digital twin systems. Nevertheless, their exploration capabilities in indoor digital-twin scenes remain limited. Dense semantic maps incur substantial maintenance costs and exhibit poor adaptability to dynamic environments. Furthermore, current navigation processes lack mechanisms for assessing and revisiting semantic relevance, leading to redundant exploration and inefficient resource utilization. Finally, environmental changes often result in semantic anchor degradation, with existing methods lacking effective incremental update and self-evolution mechanisms to maintain long-term consistency. To address these challenges, this paper proposes a multimodal, lightweight mapping, and revisitable self-evolving exploration agent (LRSA) for indoor digital-twin scenarios. The framework consists of three key modules: a multi-scale structured environment modeling process based on sparse semantic mapping, a large-model-driven memory-intelligence system that maintains an “area-object-feature” relational network, and a perceptual-memory self-evolution mechanism that marks inactive objects upon navigation failure and uses a memory large language model (MemoLLM) to generate and iteratively refine candidate search strategies for dynamic memory updating. Experiments conducted on the HM3D indoor dataset and the TurtleBot3 platform demonstrate that LRSA achieves a Success Rate (SR) of 55.2% and a Success weighted by Path Length (SPL) of 30.5%, surpassing state-of-the-art baselines such as ZSON, PSL, and PixNav. The proposed framework eliminates reliance on dense maps, enabling lightweight sparse modeling with reduced computational cost, synergistic memory-intelligence semantic reasoning, and self-adaptive evolution in dynamic environments. Its contributions enhance the robustness and efficiency of embodied agents’ open-world navigation and provide real-time, long-term semantic updating for urban digital twin systems, supporting autonomous “self-perception, self-evolution, and self-maintenance.”

Research in many fields has shown that transfer learning (TL) is well-suited to improve the performance of deep learning (DL) models in datasets with small numbers of samples. This empirical success has triggered interest in the application of TL to cognitive decoding analyses with functional neuroimaging data. Here, we systematically evaluate TL for the application of DL models to the decoding of cognitive states (e.g., viewing images of faces or houses) from whole-brain functional Magnetic Resonance Imaging (fMRI) data. We first pre-train two DL architectures on a large, public fMRI dataset and subsequently evaluate their performance in an independent experimental task and a fully independent dataset. The pre-trained DL models consistently achieve higher decoding accuracies and generally require less training time and data than model variants that were not pre-trained, while also outperforming linear baseline models trained from scratch, clearly underlining the benefits of pre-training. We demonstrate that these benefits arise from the ability of the pre-trained models to reuse many of their learned features when training with new data, providing deeper insights into the mechanisms giving rise to the benefits of pre-training. Yet, we also surface nuanced challenges for whole-brain cognitive decoding with DL models when interpreting the decoding decisions of the pre-trained models, as these have learned to utilize the fMRI data in unforeseen and counterintuitive ways to identify individual cognitive states.

Autonomous decision-making in dynamic, uncertain environments (such as robotic navigation, self-driving vehicles, and smart grid management) requires integrating powerful perception with adaptive planning. We propose a hybrid framework that combines deep learning for perceptual representation and reinforcement learning (RL) for sequential control. The framework uses convolutional or recurrent neural networks to process high-dimensional sensory data (e.g., camera or sensor arrays) and outputs features or latent states, which are then fed into an RL algorithm (such as Deep Q-Network or Actor-Critic) to learn optimal policies. We detail the architecture, including neural-network-based value/policy approximation and experience replay mechanisms, and describe our simulated testbed (e.g. CARLA urban driving simulator or OpenAI Gym environments). Experiments in a simulated dynamic scenario demonstrate that our hybrid approach significantly outperforms standard tabular RL and pure deep-learning baselines. We provide training curves, comparative tables of performance metrics, and discuss the implications for real-world deployment. Key contributions include: a conceptual integration of deep representation learning with RL for dynamic control; a detailed methodology for training in simulation; empirical results illustrating learning progress and robustness.

This chapter presents a novel application for wireless technology to assist visually impaired people. As an alternative to the medical model of rehabilitation, the information explosion era provides the foundation for a technological solution to lead the visually impaired to more independent lives in the community by minimizing the obstacles of living. A “SmartGuide” caregiver monitoring system is built as a standalone portable handheld device linked. The objective of this system is to assist blind and low vision people to walk around independently especially in dynamic changing environments. Navigation assistance is accomplished by providing speech guidance on how to move to a particular location. The system delivers dynamic environmental information to lead the visually impaired to more independent lives in the community by minimizing the obstacles of living. Information of changing environments such as road blockage, road closure, and intelligent navigation aids is provided to the user in order to guide the user safely to his or her destination. This system also includes a camera sensor network to enhance monitoring capabilities for an extra level of security and reliability.

Objective: To design a deep transfer learning framework for modeling fluence map predictions for stereotactic body radiation therapy (SBRT) of adrenal cancer and similar sites that usually have a small number of cases. Approach: We developed a transfer learning framework for adrenal SBRT planning that leverages knowledge in a pancreas SBRT planning model. Treatment plans from the two sites had different dose prescriptions and beam settings but both prioritized gastrointestinal sparing. A base framework was first trained with 100 pancreas cases. This framework consists of two convolutional neural networks (CNN), which predict individual beam doses (BD-CNN) and fluence maps (FM-CNN) sequentially for 9-beam intensity-modulated radiation therapy (IMRT) plans. Forty-five adrenal plans were split into training/validation/test sets with the ratio of 20/10/15. The base BD-CNN was re-trained with transfer learning using 5/10/15/20 adrenal training cases to produce multiple candidate adrenal BD-CNN models. The base FM-CNN was directly used for adrenal cases. The deep learning (DL) plans were evaluated by several clinically relevant dosimetric endpoints, producing a percentage score relative to the clinical plans. Main results: Transfer learning significantly reduced the number of training cases and training time needed to train such a DL framework. The adrenal transfer learning model trained with 5/10/15/20 cases achieved validation plan scores of 85.4/91.2/90.7/89.4, suggesting that model performance saturated with 10 training cases. Meanwhile, a model using all 20 adrenal training cases without transfer learning only scored 80.5. For the final test set, the 5/10/15/20-case models achieved scores of 73.5/75.3/78.9/83.3. Significance: It is feasible to use deep transfer learning to train an IMRT fluence prediction framework. This technique could adapt to different dose prescriptions and beam configurations. This framework potentially enables DL modeling for clinical sites that have a limited dataset, either due to few cases or due to rapid technology evolution.

The increasing integration of mobile robots in various industries necessitates efficient navigation strategies amidst dynamic environments. Path planning plays a crucial role in guiding mobile robots from their starting points to target destinations, contributing to automation and enhancing human-robot collaboration. This study focuses on devising a tailored path-planning approach for a fleet of mobile robots to navigate through dynamic obstacles and reach designated trajectories efficiently. Leveraging particle swarm optimization (PSO), our methodology optimizes the path while considering real-time environmental changes. We present a simulation-based implementation of the algorithm, where each robot maintains position, velocity, cost, and personal best information to converge towards the global optimal solution. Different obstacles consist of circles, squares, rectangles, and triangles with various colors and five handle-points used. Our findings demonstrate that PSO achieves a global best cost of 5.1017, indicative of the most efficient path, minimizing overall distance traveled.