PAPerpetual Hope Akwensi

Perpetual Hope Akwensi

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Perpetual Hope Akwensi has a total of 21 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Using NeRFs for UAV-based 3D reconstruction of complex scenes: A comparison to MVS

    by Frederik Schulte, Perpetual Akwensi, Lukas Winiwarter

    High-resolution 3D documentation of cultural heritage sites is essential for their preservation. While terrestrial laser scanning (TLS) remains the gold standard, it is often cost-intensive compared to photogrammetry. This study evaluates three image-based reconstruction techniques, Multi-View Stereo (MVS), Neural Radiance Fields (NeRF) and 3D Gaussian Splatting (3DGS), by applying them to a complex scene featuring a chapel and its surrounding vegetation, sensed from an uncrewed aerial vehicle (UAV). A hybrid TLS/MVS model provides a high-accuracy reference. Using identical interior and exterior camera parameters of the 105 UAV-acquired images, we generate dense point clouds with all methods and assess geometric accuracy and completeness using the M3C2 algorithm. Results show that MVS achieves superior accuracy (standard deviation of all M3C2 distances: MVS = 0.11 m, NeRF = 0.15 m), whereas NeRF attains up to 20% higher completeness, particularly in low-texture and vegetation-occluded regions. The 3DGS point cloud was deemed too sparse and was therefore not used for further analysis. The study highlights the potential of NeRFs to recover partially occluded or sparsely textured geometries that are challenging for MVS and suggests a complementary use of both approaches for cost-efficient documentation of cultural heritage.

  • Open Access

    Sub-daily Probabilistic Forecasting of High-Energy Seismic Events in Underground Mining Using Deep Temporal Classifiers: A Case Study at West African Gold Mine

    by Clara Akalanya Abuntori, Perpetual Hope Akwensi, Emmanuel D. Sunkari, Bright Tsidi Akuinor

    This study applies deep temporal classifiers to probabilistic seismic hazard assessment at a West African underground hard-rock gold mine. The analysis evaluates five forecasting horizons (1, 3, 6, 12, and 24 hours) across six model architectures, that is Gated Recurrent Unit (GRU), Transformer, Logistic Regression (LR), Gradient Boosted Machines (GBM), Long Short-Term Memory (LSTM), and Temporal Convolutional Networks (TCN). Training of the models is performed based on weighted binary cross-entropy and evaluation is done using ROC-AUC, PR-AUC, Brier score, Expected Calibration Error (ECE), partial area under the curve (pAUC) for low false positive rate, Matthews Correlation Coefficient (MCC) and balanced class accuracy. Confidence intervals are computed using the Moving Block Bootstrap (MBB; block size b = 51; 2,000 iterations) procedure which accounts for the autocorrelation of seismic events in time. The GRU obtains a maximum ROC-AUC score of 0.727 after 12 hours whereas the Transformer network obtains 0.941 at 24 hours. In the operationally important 3-hour horizon, both deep learning architectures provide ROC-AUC scores close to 0.64 and PR-AUC scores close to 0.74, outperforming all the shallow baselines. Integrated Gradients and GradientSHAP attributions at 3 hours show that the inter-event time (Δt) is the key predictive feature in both the models, corroborating the Omori-type short memory clustering hypothesis. Threshold sensitivity analysis at the 90th, 95th and 97.5th percentile of high energy events shows that the 95th percentile threshold gives the best stability of the discrimination task.

  • Open Access

    CENS: A Coverage-Efficient Pixel Sampling Strategy for Enhancing NeRF-Generated Point Cloud Fidelity

    by Perpetual Hope Akwensi, Frederik Schulte, Lukas Winiwarter

    Many geospatial workflows critically depend on high-fidelity 3D point clouds for applications such as change detection, orthophoto generation, and modeling. However, NeRF-generated point clouds often suffer from sampling inefficiencies inherent in the predominant random pixel sampling approach. We identify spatial redundancy as one such inefficiency: random sampling has the inevitable consequence of sampling large, low-texture patches more frequently than detailed, high-frequency textured regions. As a result, low-texture areas tend to be oversampled and other pixels remain unsampled – regardless of their importance to the reconstruction task. To overcome this, we propose CENS (Coverage-Efficient Non-Redundant Sampling), a deterministic pixel sampling strategy that maximizes spatial coverage, eliminates intra-image sample repetition, and ensures reproducibility via structured initialization. Evaluated on the Jamtal valley dataset, CENS achieves comparable geometric accuracy (cloud-to-mesh (C2M) distances: μ = - 0.0027 vs. -0.0011 m; σ = 0.027 vs. 0.028 m) using 50% fewer training steps (11,232 vs. 22,464), while yielding 28.2% more points, higher orthophoto fidelity, and improved point cloud completeness. Beyond CENS, we also explored NeRFs for ALS point cloud simulation, achieving realistic occlusion patterns and accuracy within UAV photogrammetry standards (VRMSE = 24 mm; HRMSE = 17 mm). Crucially, CENS positions NeRFs as a scalable, practical solution for geospatial point cloud and orthophoto generation, advancing them toward real-world mapping workflows, and integrates seamlessly into NeRFStudio.

  • Open Access

    Enhancing Multi-Temporal 3D Reconstruction of River Corridor Dynamics with Deterministic NeRF Sampling Strategies

    by Perpetual Akwensi, Frederik Schulte, Lukas Winiwarter

    Effective monitoring of hydro-geomorphological processes such as sediment displacement and channel migration using UAV-captured RGB images requires accurate, temporally consistent 3D reconstructions that enable the estimation of height/volume changes across multiple acquisitions epochs, despite variations in imaging conditions, flight paths, and camera geometries. Neural Radiance Fields (NeRF) offer a powerful framework for reconstructing complex fluvial environments from multi-view imagery. However, standard NeRF training relies on equal‑probability random pixel sampling, a strategy that introduces stochasticity into the reconstruction process and undermines the temporal repeatability required for change detection. When separate NeRFs are trained for different epochs the random sampling of pixels – combined with differing camera poses – leads to inconsistent ray distributions and surface sampling, and reduced comparability of resulting reconstructions. These effects are particularly detrimental in dynamic river corridors where low-texture surfaces, water reflections and surface motions already limit reconstruction coverage and stability.We propose Coverage-Efficient Non-redundant Sampling (CENS) as a grid-based deterministic alternative to random pixel sampling in NeRF training, combined with a pose-consistent inference framework, to improve temporal reconstruction repeatability for hydro-geomorphic change analysis. Instead of randomly sampling pixels per iteration, we impose a spatially deterministic sampling grid that ensures intra-epoch spatial coverage and consistency. To address the challenge of differing camera poses between epochs, we introduce a cross‑epoch evaluation strategy in which all epochs are queried using a shared set of reference camera poses – either drawn from one epoch’s camera set or defined as a synthetic virtual camera lattice – after alignment to a common world coordinate frame. This enables pose‑consistent rendering across time even when the original acquisition geometries differ.Rendered depth maps from these shared viewpoints are then projected onto a fixed world-space grid, enabling cell-wise comparison of height and volume change. This two-layer determinism—fixed sampling during training and fixed camera/grid sampling during inference—decouples temporal analysis from the stochastic and pose‑dependent variability inherent in standard NeRF pipelines.We will evaluate the method on multi‑epoch UAV imagery of an active river reach, comparing reconstructions trained with traditional random pixel sampling against those produced using the proposed CENS approach. With improved intra-epoch spatial coverage and geometric stability, combined with pose-consistent inference, we anticipate that the proposed approach will yield more coherent inter-epoch estimates of sediment elevation, channel-bed change and/or bank erosion, with lower uncertainty in low-textured or reflective regions. We expect corresponding 3D points from different epochs to align more consistently in world space, enabling more reliable detection of subtle geomorphic adjustments like bar migration, scour-fill cycles, and/or sediment redistribution.This study aims to demonstrate that sampling strategy and inference geometry are critical, yet previously overlooked, determinants of temporal repeatability in NeRF-based hydro-geomorphological monitoring. By replacing stochastic pixel sampling with deterministic grids and enforcing shared inference poses across epochs, we introduce a simple but potentially powerful modification that may enhance the reliability of NeRF-derived height/volume change estimates in dynamic river environments. The approach could open new possibilities for long‑term environmental and hydro-geomorphological research where spatiotemporal consistency is essential.

  • Open Access

    An enhanced network for extracting tunnel lining defects using transformer encoder and aggregate decoder

    by Bo Guo, Zhihai Huang, Haitao Luo, Perpetual Hope Akwensi, Ruisheng Wang, Bo Huang, Tsz Nam Chan


  • Performance evaluation of artificial neural networks for natural terrain classification

    by Perpetual Hope Akwensi, Eric Thompson Brantson, Johanna Ngula Niipele, Yao Yevenyo Ziggah


  • Reconstruction of Power Pylons From LiDAR Point Clouds Based on Structural Segmentation and Parameter Estimation

    by Hui Wang, Teng Hu, Zhen Wang, Zhizhong Kang, Perpetual Hope Akwensi, Juntao Yang

    The reconstruction of 3-D models of power pylons from light detection and ranging (LiDAR) data plays an important role in power transmission safety. However, accurate reconstruction of power pylon models still faces challenges, e.g., complex structures, missing data, and occlusion. In this letter, a novel four-component segmentation method is proposed for reconstructing power pylon models. In the proposed method, the pylon components of the pylon head, pylon body, cross-arms, and pedestal are first defined in terms of the common features and functionality of each component. Then, these four components are each segmented and identified based on their position and shape features from the raw point cloud. An improved approach based on Metropolis–Hastings sampling and a simulated annealing algorithm is proposed to estimate the model parameters. Based on the estimated parameters, the 3-D shapes of the individual components are reconstructed and stacked to form the whole pylon model. Experimental results show that our methods are able to reconstruct pylons with complex-shaped heads and multiple cross-arms, with an average reconstruction error of less than 0.3 m. Compared with the standard Metropolis–Hastings algorithm with annealing, the parameter estimation process in our strategy improves the computational efficiency by 7.54%.


  • A Label-Constraint Building Roof Detection Method From Airborne LiDAR Point Clouds

    by Juntao Yang, Zhizhong Kang, Perpetual Hope Akwensi

    Airborne light detection and ranging (LiDAR) point clouds have become growingly popular as a reliable data source for 3-D digital building model reconstruction. Therefore, we develop a label-constraint approach for automatically detecting building roofs using airborne LiDAR point clouds and multispectral images, where the label information is introduced in both the discriminative feature space generation and the detection procedure. To obtain a robust and highly discriminative descriptor, a supervised sparse coding-enhanced bag of visual word (SC-BOVW) model based on a learned discriminative dictionary is used to encode local geometric and spectral information within each super-voxel into high-level semantic representation, which is then fed into a support vector machine (SVM) classifier for distinguishing buildings from others. Additionally, a graph cut-based procedure is used as a postprocessing step to guarantee the spatial consistency in detection results. Experiments were conducted on the International Society for Photogrammetry and Remote Sensing (ISPRS) benchmark data sets. Results indicate that the proposed method is accurate and efficient in terms of building roof region detection. Moreover, the proposed method is superior to other existing methods with average differences in recall of 2.23%, precision of 0.28% and quality of 1.99%.


  • Open Access

    An Individual Tree Segmentation Method Based on Watershed Algorithm and Three-Dimensional Spatial Distribution Analysis From Airborne LiDAR Point Clouds

    by Juntao Yang, Zhizhong Kang, Sai Cheng, Zhou Yang, Perpetual Hope Akwensi

    Accurate individual tree segmentation is an important basis for the subsequent calculation and analysis of forestry parameters. However, rasterized Canopy Height Model (CHM)-based methods often suffer from 3D information loss due to the interpolation operation. Therefore, this paper proposes an individual tree segmentation method based on the marker-controlled watershed algorithm and 3D spatial distribution analysis from airborne LiDAR point clouds. First, based on the potential tree apices derived from the local maxima filtering, the marker-controlled watershed segmentation algorithm is conducted to obtain the coarse point clusters. Then, within the Principal Component Analysis (PCA)-defined local coordinate reference framework, a multi-directional 3D spatial profile analysis is performed on each point cluster to refine the potential tree apex positions. Finally, the refined potential tree apex positions are used as a prior of K-means clustering to achieve the coarse-to-fine individual tree segmentation. Comparative experiments were conducted on the public NEWFOR dataset to evaluate the proposed method. Results indicate that the proposed method is efficient and robust for segmenting individual trees.


  • Fisher Vector Encoding of Supervoxel-Based Features for Airborne LiDAR Data Classification

    by Perpetual Hope Akwensi, Zhizhong Kang, Juntao Yang

    Point cloud feature extraction as a classification task is crucial in maximizing the efficient downstream applicability of raw point clouds. With the goal of learning optimum features for efficient classification of a multi-class point cloud for downstream applications, this letter presents a supervoxel-Fisher vector (FV)-based approach for airborne light detection and ranging (LiDAR) data classification. In our approach, FV encoding is implemented to deduce compact global descriptors from aggregated supervoxels to establish a more descriptive and discriminative representation, transforming the low-level visual features into high-level semantic features. As a result, the proposed approach combines local and global feature properties through the quantization and aggregation of higher order statistics to harnesses their combined advantages for producing good classification results. Experiments were conducted on the international society for photogrammetry and remote sensing 3-D semantic labeling benchmark data set. Results indicate that the proposed approach is robust and efficient, attained the third best position with an overall accuracy of 81.79%, and ranked first with an F₁-score of 72.31%.