arXiv subjects

M. P. B. Ekanayake

Publications and source records attributed to M. P. B. Ekanayake.

13 recordsLinked to original sources

arXiv2026

Reflectance Multispectral Imaging for Soil Composition Estimation and USDA Texture Classification

Soil texture is a foundational attribute that governs water availability and erosion in agriculture, as well as load bearing capacity, deformation response, and shrink-swell risk in geotechnical engineering. Yet texture is still typically determined by slow and labour intensive laboratory particle size tests, while many sensing alternatives are either costly or too coarse to support routine field scale deployment. This paper proposes a robust and field deployable multispectral imaging (MSI) system and machine learning framework for predicting soil composition and the United States Department of Agriculture (USDA) texture classes. The proposed system uses a cost effective in-house MSI device operating from 365 nm to 940 nm to capture thirteen spectral bands, which effectively capture the spectral properties of soil texture. Regression models use the captured spectral properties to estimate clay, silt, and sand percentages, while a direct classifier predicts one of the twelve USDA textural classes. Indirect classification is obtained by mapping the regressed compositions to texture classes via the USDA soil texture triangle. The framework is evaluated on mixture data by mixing clay, silt, and sand in varying proportions, using the USDA classification triangle as a basis. Experimental results show that the proposed approach achieves a coefficient of determination R^2 up to 0.99 for composition prediction and over 99% accuracy for texture classification. These findings indicate that MSI combined with data-driven modeling can provide accurate, non-destructive, and field deployable soil texture characterization suitable for geotechnical screening and precision agriculture.

cs.CV↗

arXiv2025

I-Diff: Structural Regularization for High-Fidelity Diffusion Models

Denoising Diffusion Probabilistic Models (DDPMs) have significantly advanced generative AI, achieving impressive results in high-quality image and data generation. However, enhancing fidelity without compromising semantic content remains a key challenge in the field. Recent diffusion research in multiple disciplines has introduced objectives and architectural refinements that tighten the match between generated and real data distributions, yielding higher fidelity than earlier generative frameworks. Multi-stage architectures, physics-guided modeling, semantic conditioning, and rarity-aware generation are some of the explored works to achieve this task. However, the integration of structural information of the data distribution into DDPM has largely been unexplored. The conventional DDPM framework relies solely on the $L^2$ norm between the additive and predicted noise to generate new data distributions. We introduce I-Diff, an improved version of DDPM that incorporates a carefully designed regularizer, effectively enabling the model to encode structural information, thereby preserving the inherent fidelity of the data distribution. The proposed approach is validated through extensive experiments on DDPM, Improved DDPM and Latent Diffusion Model across multiple datasets. Empirical results demonstrate significant improvements in fidelity (Density and Precision increase 10% and 37% in CIFAR-100 dataset respectively) across other tested datasets. These results highlight the effectiveness of our method in enhancing the fidelity of the generated data. Notably, improvements across different models highlight the model-agnostic nature of our proposed method in the wider field of generative AI.

cs.LG↗

arXiv2025

Computer Vision based group activity detection and action spotting

Group activity detection in multi-person scenes is challenging due to complex human interactions, occlusions, and variations in appearance over time. This work presents a computer vision based framework for group activity recognition and action spotting using a combination of deep learning models and graph based relational reasoning. The system first applies Mask R-CNN to obtain accurate actor localization through bounding boxes and instance masks. Multiple backbone networks, including Inception V3, MobileNet, and VGG16, are used to extract feature maps, and RoIAlign is applied to preserve spatial alignment when generating actor specific features. The mask information is then fused with the feature maps to obtain refined masked feature representations for each actor. To model interactions between individuals, we construct Actor Relation Graphs that encode appearance similarity and positional relations using methods such as normalized cross correlation, sum of absolute differences, and dot product. Graph Convolutional Networks operate on these graphs to reason about relationships and predict both individual actions and group level activities. Experiments on the Collective Activity dataset demonstrate that the combination of mask based feature refinement, robust similarity search, and graph neural network reasoning leads to improved recognition performance across both crowded and non crowded scenarios. This approach highlights the potential of integrating segmentation, feature extraction, and relational graph reasoning for complex video understanding tasks.

cs.CV↗

arXiv2025

Dual mode multispectral imaging system for food and agricultural product quality estimation

Multispectral imaging coupled with Artificial Intelligence, Machine Learning and Signal Processing techniques work as a feasible alternative for laboratory testing, especially in food quality control. Most of the recent related research has been focused on reflectance multispectral imaging but a system with both reflectance, transmittance capabilities would be ideal for a wide array of specimen types including solid and liquid samples. In this paper, a device which includes a dedicated reflectance mode and a dedicated transmittance mode is proposed. Dual mode operation where fast switching between two modes is facilitated. An innovative merged mode is introduced in which both reflectance and transmittance information of a specimen are combined to form a higher dimensional dataset with more features. Spatial and temporal variations of measurements are analyzed to ensure the quality of measurements. The concept is validated using a standard color palette and specific case studies are done for standard food samples such as turmeric powder and coconut oil proving the validity of proposed contributions. The classification accuracy of standard color palette testing was over 90% and the accuracy of coconut oil adulteration was over 95%. while the merged mode was able to provide the best accuracy of 99% for the turmeric adulteration. A linear functional mapping was done for coconut oil adulteration with an R2 value of 0.9558.

eess.SP↗

arXiv2024

Transformer based Endmember Fusion with Spatial Context for Hyperspectral Unmixing

In recent years, transformer-based deep learning networks have gained popularity in Hyperspectral (HS) unmixing applications due to their superior performance. The attention mechanism within transformers facilitates input-dependent weighting and enhances contextual awareness during training. Drawing inspiration from this, we propose a novel attention-based Hyperspectral Unmixing algorithm called Transformer-based Endmember Fusion with Spatial Context for Hyperspectral Unmixing (FusionNet). This network leverages an ensemble of endmembers for initial guidance, effectively addressing the issue of relying on a single initialization. This approach helps avoid suboptimal results that many algorithms encounter due to their dependence on a singular starting point. The FusionNet incorporates a Pixel Contextualizer (PC), introducing contextual awareness into abundance prediction by considering neighborhood pixels. Unlike Convolutional Neural Networks (CNNs) and traditional Transformer-based approaches, which are constrained by specific kernel or window shapes, the Fusion network offers flexibility in choosing any arbitrary configuration of the neighborhood. We conducted a comparative analysis between the FusionNet algorithm and eight state-of-the-art algorithms using three widely recognized real datasets and one synthetic dataset. The results demonstrate that FusionNet offers competitive performance compared to the other algorithms.

eess.IV↗

arXiv2024

A Novel Algorithm for Digital Lithological Mapping-Case Studies in Sri Lanka's Mineral Exploration

Conventional manual lithological mapping (MLM) through field surveys are resource-extensive and time-consuming. Digital lithological mapping (DLM), harnessing remotely sensed spectral imaging techniques, provides an effective strategy to streamline target locations for MLM or an efficient alternative to MLM. DLM relies on laboratory-generated generic end-member signatures of minerals for spectral analysis. Thus, the accuracy of DLM may be limited due to the presence of site-specific impurities. A strategy, based on a hybrid machine-learning and signal-processing algorithm, is proposed in this paper to tackle this problem of site-specific impurities. In addition, a soil pixel alignment strategy is proposed here to visualize the relative purity of the target minerals. The proposed methodologies are validated via case studies for mapping of Limestone deposits in Jaffna, Ilmenite deposits in Pulmoddai and Mannar, and Montmorillonite deposits in Murunkan, Sri Lanka. The results of satellite-based spectral imaging analysis were corroborated with X-ray diffraction (XRD) and Magnetic Separation (MS) analysis of soil samples collected from those sites via field surveys. There exists a good correspondence between the relative availability of the minerals with the XRD and MS results. In particular, correlation coefficients of 0.8115 and 0.9853 were found for the sites in Pulmoddai and Jaffna respectively.

eess.IV↗

arXiv2023

Laser Ranging Based Intelligent System for Unknown Environment Mapping

This work describes the implementation of a simple and computationally efficient Intelligent Navigation System (INS) for autonomous systems used in areas where human access is impossible. The system uses Laser Range Finder (LRF) readings as input, making it suitable for mobile platform implementation. The INS pre-processes the LRF readings to remove noise and determines an obstacle-free path for mapping. The system's localization method uses a similarity transform and particle filter. The system was tested in artificially generated environments and emulated in real-time with real-environment data. The system was then implemented in a Raspberry Pi3 on a 3WD Omni-directional mobile platform and tested in real environments. The system was able to generate an accurate 2D map of the area. The proposed methodology was shown to be efficient through a comparative analysis of execution time.

cs.RO↗

arXiv2023

Object Dimension Extraction for Environment Mapping with Low Cost Cameras Fused with Laser Ranging

It is essential to have a method to map an unknown terrain for various applications. For places where human access is not possible, a method should be proposed to identify the environment. Exploration, disaster relief, transportation and many other purposes would be convenient if a map of the environment is available. Replicating the human vision system using stereo cameras would be an optimum solution. In this work, we have used laser ranging based technique fused with stereo cameras to extract dimension of objects for mapping. The distortions were calibrated using mathematical model of the camera. By means of Semi Global Block Matching [1] disparity map was generated and reduces the noise using novel noise reduction method of disparity map by dilation. The Data from the Laser Range Finder (LRF) and noise reduced vision data has been used to identify the object parameters.

cs.CV↗

arXiv2022

Transmittance Multispectral Imaging for Reheated Coconut Oil Differentiation

Oil reheating has a significant impact on global health due to its extensive consumption, especially in South Asia, and severe health risks. Nevertheless, food image analysis using multispectral imaging systems(MISs) has not been applied to oil reheating analysis despite their vast application in rapid food quality screening. To that end, the paper discusses the application of a low-cost MSI to estimate the 'reheat cycle count classes' (number of times an oil sample is recursively heated) and identify 'critical classes' at which substantial changes in the oil sample have materialized. Firstly, the reheat cycle count class is estimated with Bhattacharyya distance between the reheated and a pure oil sample as the input. The classification was performed using a support vector machine classifier that resulted in an accuracy of 83.34 % for reheat cycle count identification. Subsequently, an unsupervised clustering procedure was introduced using a modified spectral clustering (SC) algorithm to distinguish critical classes under reheating. In addition, laboratory experiments were performed to ascertain the ramifications of the reheating process with a chemical analysis. The chemical analysis of the coconut oil samples used in the experiment coincided with the chemical analysis results and was statistically significant (p < 0.05). Accordingly, the proposed work closes the gap for using multispectral imaging for oil reheating and proposes a novel algorithm for unsupervised detection of critical property changes in the oil. Hence, the proposed research work is significant in its practical implications, contribution to food image analysis, and unsupervised classification mechanisms.

eess.IV↗

arXiv2021

Single-target mineral detection with site-specific endmember extraction for survey points identification: A case study of Jaffna, Sri Lanka

As field surveys used for manual lithological mapping are costly and time-consuming, digital lithological mapping (DLM) that utilizes remotely sensed spectral imaging provides a viable and economical alternative. Generally, DLM has been performed using spectral imaging with the use of laboratory-generated generic endmember signatures. To that end, this paper proposes generating a single-target abundance mineral map for DLM, where the generated map can further be used as a guide for the selection or avoidance of a field survey. For that, a stochastic cancellation-based methodology was used to generate a site-specific endemic signature for the mineral in concern to reduce the inclusive nature otherwise present in DLM. Furthermore, a soil pixel alignment strategy to visualize the relative purity level of the target mineral has been introduced in the proposed work. Then, for the method validation, mapping of limestone deposits in the Jaffna peninsula of Sri Lanka was conducted as the case study using satellite-based spectral imaging as the input. It was observed that despite the low signal-to-noise ratio of the input hyperspectral data the proposed methodology was able to robustly extract the rich information contained in the input data. Further, a field survey was conducted to collect soil samples of four sites chosen by the proposed DLM from the Jaffna peninsula as an algorithm validation and to demonstrate the application of the proposed solution. The proposed abundance threshold of 0.1 coincided with the industrial standard X-ray diffraction (XRD) threshold of 5% for the mineral presence. The results of the XRD test validated the use of the algorithm in the selection of sites to be surveyed, hence could avoid conducting a costly field survey on the assumption of the existence of a mineral.

physics.geo-ph↗

arXiv2021

Constrained Nonnegative Matrix Factorization for Blind Hyperspectral Unmixing incorporating Endmember Independence

Hyperspectral unmixing (HU) has become an important technique in exploiting hyperspectral data since it decomposes a mixed pixel into a collection of endmembers weighted by fractional abundances. The endmembers of a hyperspectral image (HSI) are more likely to be generated by independent sources and be mixed in a macroscopic degree before arriving at the sensor element of the imaging spectrometer as mixed spectra. Over the past few decades, many attempts have focused on imposing auxiliary constraints on the conventional nonnegative matrix factorization (NMF) framework in order to effectively unmix these mixed spectra. As a promising step toward finding an optimum constraint to extract endmembers, this paper presents a novel blind HU algorithm, referred to as Kurtosis-based Smooth Nonnegative Matrix Factorization (KbSNMF) which incorporates a novel constraint based on the statistical independence of the probability density functions of endmember spectra. Imposing this constraint on the conventional NMF framework promotes the extraction of independent endmembers while further enhancing the parts-based representation of data. Experiments conducted on diverse synthetic HSI datasets (with numerous numbers of endmembers, spectral bands, pixels, and noise levels) and three standard real HSI datasets demonstrate the validity of the proposed KbSNMF algorithm compared to several state-of-the-art NMF-based HU baselines. The proposed algorithm exhibits superior performance especially in terms of extracting endmember spectra from hyperspectral data; therefore, it could uplift the performance of recent deep learning HU methods which utilize the endmember spectra as supervisory input data for abundance extraction.

eess.IV↗

arXiv2020

Quantitative Assessment of Adulteration and Reuse of Coconut Oil Using Transmittance Multispectral Imaging

Coconut oil known for its wide range of uses is often adulterated with other edible oils. Repeated use of coconut oil in food preparation could lead to many health issues. Existing methods available for evaluating quality of oil are laborious and time consuming. Therefore, we propose an imaging system hardware and image processing-based algorithm to estimate the adulteration of coconut oil with palm oil as the adulterant. A clear functional relationship between adulteration level and Bhattacharyya distance was observed as R2 = 0.9876 on the training samples. Thereafter, another algorithm is proposed to develop a spectral-clustering based classifier to determine the effect of reheat and reuse of coconut oil. Distinct clusters were obtained for different levels of reheated oil classes and the classification was performed with an accuracy of 0.983 on training samples. Further, the input images for the proposed algorithms were generated using an in-house developed transmittance based multispectral imaging system.

eess.SP↗

arXiv2020

Coordinated PV re-phasing: a novel method to maximize renewable energy integration in LV networks by mitigating network unbalances

As combating climate change has become a top priority and as many countries are taking steps to make their power generation sustainable, there is a marked increase in the use of renewable energy sources (RESs) for electricity generation. Among these RESs, solar photovoltaics (PV) is one of the most popular sources of energy connected to LV distribution networks. With the greater integration of solar PV into LV distribution networks, utility providers impose caps to solar penetration in order to operate their network safely and within acceptable norms. One parameter that restricts solar PV penetration is unbalances created by loads and single-phase rooftop schemes connected to LV distribution grids. In this paper, a novel method is proposed to mitigate voltage unbalance in LV distribution grids by optimally re-phasing grid-connected rooftop PV systems. A modified version of the discrete bacterial foraging optimization algorithm (DBFOA) is introduced as the optimization technique to minimize the overall voltage unbalance of the network as the objective function, subjected to various network and operating parameters. The impact of utilizing the proposed PV re-phasing technique as opposed to a fixed phase configuration are compared based on overall voltage unbalance, which was observed hourly throughout the day. The case studies show that the proposed approach can significantly mitigate the overall voltage unbalance during the daytime and can facilitate to increase the usable PV capacity of the considered network by 77%.

eess.SY↗