Roshan Ragel

Romiyal George, Sathiyamohan Nishankar, Selvarajah Thuseethan et al.

Vision Transformers (ViTs) have demonstrated strong representation capability in image classification. However, their quadratic self-attention complexity and large parameter counts limit deployment on resource-constrained mobile and edge devices. This paper introduces UtVAA, an ultra-tiny Vision Transformer architecture designed for efficient visual recognition under strict computational budgets. It incorporates a novel Affix Attention block that combines depthwise-pointwise local feature extraction, linear self-attention, coordinate attention for spatial dependency modelling, and a lightweight ternary fusion strategy to integrate local and global representations. In addition, Dilated Bottleneck blocks expand the receptive field using dilated depthwise separable convolutions while maintaining low FLOPs and stable optimisation through residual connections. UtVAA is implemented in scalable Tiny, Medium, and Large variants, with the smallest model containing 204.67K parameters and 53.95M FLOPs. Experimental results on CIFAR-10, CIFAR-100, PlantVillage-Tomato and SLIF-Tomato datasets show that UtVAA achieves competitive accuracy within a sub-million-parameter regime. Overall, the results demonstrate that transformer-based vision models can be redesigned into ultra-tiny architectures without significant loss in discriminative performance, making UtVAA suitable for mobile and edge deployment. Code is available at https://github.com/romiyal/UtVAA

Geesara Prathap, Titus Nanda Kumara, Roshan Ragel

Recognizing sEMG (Surface Electromyography) signals belonging to a particular action (e.g., lateral arm raise) automatically is a challenging task as EMG signals themselves have a lot of variation even for the same action due to several factors. To overcome this issue, there should be a proper separation which indicates similar patterns repetitively for a particular action in raw signals. A repetitive pattern is not always matched because the same action can be carried out with different time duration. Thus, a depth sensor (Kinect) was used for pattern identification where three joint angles were recording continuously which is clearly separable for a particular action while recording sEMG signals. To Segment out a repetitive pattern in angle data, MDTW (Moving Dynamic Time Warping) approach is introduced. This technique is allowed to retrieve suspected motion of interest from raw signals. MDTW based on DTW algorithm, but it will be moving through the whole dataset in a pre-defined manner which is capable of picking up almost all the suspected segments inside a given dataset an optimal way. Elevated bicep curl and lateral arm raise movements are taken as motions of interest to show how the proposed technique can be employed to achieve auto identification and labelling. The full implementation is available at https://github.com/GPrathap/OpenBCIPython

Kanishka Gunawardana, Sanka Peeris, Kavishka Rambukwella et al.

Spiking Neural Networks (SNNs) have gained significant attention in edge computing due to their low power consumption and computational efficiency. However, existing implementations either use conventional System on Chip (SoC) architectures that suffer from memory-processor bottlenecks, or large-scale neuromorphic hardware that is inefficient and wasteful for small-scale SNN applications. This work presents SNAP-V, a RISC-V-based neuromorphic SoC with two accelerator variants: Cerebra-S (bus-based) and Cerebra-H (Network-on-Chip (NoC)-based) which are optimized for small-scale SNN inference, integrating a RISC-V core for management tasks, with both accelerators featuring parallel processing nodes and distributed memory. Experimental results show close agreement between software and hardware inference, with an average accuracy deviation of 2.62% across multiple network configurations, and an average synaptic energy of 1.05 pJ per synaptic operation (SOP) in 45 nm CMOS technology. These results show that the proposed solution enables accurate, energy-efficient SNN inference suitable for real-time edge applications.

Ridma Jayasundara, Ishan Fernando, Adeepa Fernando et al.

Cardiovascular diseases (CVDs) are the leading cause of mortality worldwide, accounting for over 30% of global deaths according to the World Health Organization (WHO). Importantly, one-third of these deaths are preventable with timely and accurate diagnosis. The electrocardiogram (ECG), a non-invasive method for recording the electrical activity of the heart, is crucial for diagnosing CVDs. However, privacy concerns surrounding the use of patient ECG data in research have spurred interest in synthetic data, which preserves the statistical properties of real data without compromising patient confidentiality. This study explores the potential of synthetic ECG data for training deep learning models from regression to classification tasks and evaluates the feasibility of transfer learning to enhance classification performance on real ECG data. We experimented with popular deep learning models to predict four key cardiac parameters, namely, Heart Rate (HR), PR interval, QT interval, and QRS complex-using separate regression models. Subsequently, we leveraged these regression models for transfer learning to perform 5-class ECG signal classification. Our experiments systematically investigate whether transfer learning from regression to classification is viable, enabling better utilization of diverse open-access and synthetic ECG datasets. Our findings demonstrate that transfer learning from regression to classification improves classification performance, highlighting its potential to maximize the utility of available data and advance deep learning applications in this domain.

Chathura Wimalasiri, Piumal Rathnayake, Shamod Wijerathne et al.

Periodontitis, a chronic inflammatory disease causing alveolar bone loss, significantly affects oral health and quality of life. Accurate assessment of bone loss severity and pattern is critical for diagnosis and treatment planning. In this study, we propose a novel AI-based deep learning framework to automatically detect and quantify alveolar bone loss and its patterns using intraoral periapical (IOPA) radiographs. Our method combines YOLOv8 for tooth detection with Keypoint R-CNN models to identify anatomical landmarks, enabling precise calculation of bone loss severity. Additionally, YOLOv8x-seg models segment bone levels and tooth masks to determine bone loss patterns (horizontal vs. angular) via geometric analysis. Evaluated on a large, expertly annotated dataset of 1000 radiographs, our approach achieved high accuracy in detecting bone loss severity (intra-class correlation coefficient up to 0.80) and bone loss pattern classification (accuracy 87%). This automated system offers a rapid, objective, and reproducible tool for periodontal assessment, reducing reliance on subjective manual evaluation. By integrating AI into dental radiographic analysis, our framework has the potential to improve early diagnosis and personalized treatment planning for periodontitis, ultimately enhancing patient care and clinical outcomes.

Sathira Silva, Savindu Bhashitha Wannigama, Gihan Jayatilaka et al.

Holistic understanding and reasoning in 3D scenes are crucial for the success of autonomous driving systems. The evolution of 3D semantic occupancy prediction as a pretraining task for autonomous driving and robotic applications captures finer 3D details compared to traditional 3D detection methods. Vision-based 3D semantic occupancy prediction is increasingly overlooked in favor of LiDAR-based approaches, which have shown superior performance in recent years. However, we present compelling evidence that there is still potential for enhancing vision-based methods. Existing approaches predominantly focus on spatial cues such as tri-perspective view (TPV) embeddings, often overlooking temporal cues. This study introduces S2TPVFormer, a spatiotemporal transformer architecture designed to predict temporally coherent 3D semantic occupancy. By introducing temporal cues through a novel Temporal Cross-View Hybrid Attention mechanism (TCVHA), we generate Spatiotemporal TPV (S2TPV) embeddings that enhance the prior process. Experimental evaluations on the nuScenes dataset demonstrate a significant +4.1% of absolute gain in mean Intersection over Union (mIoU) for 3D semantic occupancy compared to baseline TPVFormer, validating the effectiveness of S2TPVFormer in advancing 3D scene perception.

Harshana Weligampola, Gihan Jayatilaka, Suren Sritharan et al.

Intrinsic Image Decomposition is an open problem of generating the constituents of an image. Generating reflectance and shading from a single image is a challenging task specifically when there is no ground truth. There is a lack of unsupervised learning approaches for decomposing an image into reflectance and shading using a single image. We propose a neural network architecture capable of this decomposition using physics-based parameters derived from the image. Through experimental results, we show that (a) the proposed methodology outperforms the existing deep learning-based IID techniques and (b) the derived parameters improve the efficacy significantly. We conclude with a closer analysis of the results (numerical and example images) showing several avenues for improvement.

Harshana Weligampola, Gihan Jayatilaka, Suren Sritharan et al.

Low light image enhancement is an important challenge for the development of robust computer vision algorithms. The machine learning approaches to this have been either unsupervised, supervised based on paired dataset or supervised based on unpaired dataset. This paper presents a novel deep learning pipeline that can learn from both paired and unpaired datasets. Convolution Neural Networks (CNNs) that are optimized to minimize standard loss, and Generative Adversarial Networks (GANs) that are optimized to minimize the adversarial loss are used to achieve different steps of the low light image enhancement process. Cycle consistency loss and a patched discriminator are utilized to further improve the performance. The paper also analyses the functionality and the performance of different components, hidden layers, and the entire pipeline.

MAC Jiffriya, MAC Akmal Jahan, Roshan G. Ragel

Plagiarism is known as illegal use of others' part of work or whole work as one's own in any field such as art, poetry, literature, cinema, research and other creative forms of study. Plagiarism is one of the important issues in academic and research fields and giving more concern in academic systems. The situation is even worse with the availability of ample resources on the web. This paper focuses on an effective plagiarism detection tool on identifying suitable intra-corpal plagiarism detection for text based assignments by comparing unigram, bigram, trigram of vector space model with cosine similarity measure. Manually evaluated, labelled dataset was tested using unigram, bigram and trigram vector. Even though trigram vector consumes comparatively more time, it shows better results with the labelled data. In addition, the selected trigram vector space model with cosine similarity measure is compared with tri-gram sequence matching technique with Jaccard measure. In the results, cosine similarity score shows slightly higher values than the other. Because, it focuses on giving more weight for terms that do not frequently exist in the dataset and cosine similarity measure using trigram technique is more preferable than the other. Therefore, we present our new tool and it could be used as an effective tool to evaluate text based electronic assignments and minimize the plagiarism among students.

Akmal Jahan Mac, Roshan G Ragel

Pool of knowledge available to the mankind depends on the source of learning resources, which can vary from ancient printed documents to present electronic material. The rapid conversion of material available in traditional libraries to digital form needs a significant amount of work if we are to maintain the format and the look of the electronic documents as same as their printed counterparts. Most of the printed documents contain not only characters and its formatting but also some associated non text objects such as tables, charts and graphical objects. It is challenging to detect them and to concentrate on the format preservation of the contents while reproducing them. To address this issue, we propose an algorithm using local thresholds for word space and line height to locate and extract all categories of tables from scanned document images. From the experiments performed on 298 documents, we conclude that our algorithm has an overall accuracy of about 75% in detecting tables from the scanned document images. Since the algorithm does not completely depend on rule lines, it can detect all categories of tables in a range of scanned documents with different font types, styles and sizes to extract their formatting features. Moreover, the algorithm can be applied to locate tables in multi column layouts with small modification in layout analysis. Treating tables with their existing formatting features will tremendously help the reproducing of printed documents for reprinting and updating purposes.

G. I. Gunarathna, M. A. P. Chamikara, R. G. Ragel

Character recognition techniques for printed documents are widely used for English language. However, the systems that are implemented to recognize Asian languages struggle to increase the accuracy of recognition. Among other Asian languages (such as Arabic, Tamil, Chinese), Sinhala characters are unique, mainly because they are round in shape. This unique feature makes it a challenge to extend the prevailing techniques to improve recognition of Sinhala characters. Therefore, a little attention has been given to improve the accuracy of Sinhala character recognition. A novel method, which makes use of this unique feature, could be advantageous over other methods. This paper describes the use of a fuzzy inference system to recognize Sinhala characters. Feature extraction is mainly focused on distance and intersection measurements in different directions from the center of the letter making use of the round shape of characters. The results showed an overall accuracy of 90.7% for 140 instances of letters tested, much better than similar systems.

C. B Bulumulla, R. G. Ragel

As smart phones and tablets are becoming ubiquitous and taking over as the primary choice for accessing the Internet worldwide, ensuring a secure gateway to the servers serving such devices become essential. CAPTCHAs play an important role in identifying human users in internet to prevent unauthorized bot attacks. Even though there are numerous CAPTCHA alternatives available today, there are certain drawbacks attached with each alternative, making them harder to find a general solution for the necessity of a CAPTCHA mechanism. With the advancing technology and expertise in areas such as AI, cryptography and image processing, it has come to a stage where the chase between making and breaking CAPTCHAs are even now. This has led the humans with a hard time deciphering the CAPTCHA mechanisms. In this paper, we adapt a novel CAPTCHA mechanism named as LineCAPTCHA to mobile devices. LineCAPTCHA is a new reverse Turing test based on drawing on top of Bezier curves within noisy backgrounds. The major objective of this paper is to report the implementation and evaluation of LineCAPTCHA on a mobile platform. At the same time we impose certain security standards and security aspects for establishing LineCAPTCHAs which are obtained through extensive measures. Independency from factors such as the fluency in English language, age and easily understandable nature of it inclines the usability of LineCAPTCHA. We believe that such independency will favour the main target of LineCAPTCHA, user friendliness and usability.

Janaka Alawatugoda, Darshana Jayasinghe, Roshan Ragel

Cache timing attack is a type of side channel attack where the leaking timing information due to the cache behaviour of a crypto system is used by an attacker to break the system. Advanced Encryption Standard (AES) was considered a secure encryption standard until 2005 when Daniel Bernstein claimed that the software implementation of AES is vulnerable to cache timing attack. Bernstein demonstrated a remote cache timing attack on a software implementation of AES. The original AES implementation can methodically be altered to prevent the cache timing attack by hiding the natural cache-timing pattern during the encryption while preserving its semantics. The alternations while preventing the attack should not make the implementation very slow. In this paper, we report outcomes of our experiments on designing and implementing a number of possible countermeasures.

A. Barnes, R. Fernando, K. Mettananda et al.

AES, Advanced Encryption Standard, can be considered the most widely used modern symmetric key encryption standard. To encrypt/decrypt a file using the AES algorithm, the file must undergo a set of complex computational steps. Therefore a software implementation of AES algorithm would be slow and consume large amount of time to complete. The immense increase of both stored and transferred data in the recent years had made this problem even more daunting when the need to encrypt/decrypt such data arises. As a solution to this problem, in this paper, we present an extensive study of enhancing the throughput of AES encryption algorithm by utilizing the state of the art multicore architectures. We take a sequential program that implements the AES algorithm and convert the same to run on multicore architectures with minimum effort. We implement two different parallel programmes, one with the fork system call in Linux and the other with the pthreads, the POSIX standard for threads. Later, we ran both the versions of the parallel programs on different multicore architectures and compared and analysed the throughputs between the implementations and among different architectures. The pthreads implementation outperformed in all the experiments we conducted and the best throughput obtained is around 7Gbps on a 32-core processor (the largest number of cores we had) with the pthreads implementation.

U. Herath, J. Alawatugoda, R. G. Ragel

Advanced Encryption Standard (AES) is a symmetric key encryption algorithm which is extensively used in secure electronic data transmission. When introduced, although it was tested and declared as secure, in 2005, a researcher named Bernstein claimed that it is vulnerable to side channel attacks. The cache-based timing attack is the type of side channel attack demonstrated by Bernstein, which uses the timing variation in cache hits and misses. This kind of attacks can be prevented by masking the actual timing information from the attacker. Such masking can be performed by altering the original AES software implementation while preserving its semantics. This paper presents possible software implementation level countermeasures against Bernstein's cache timing attack. Two simple software based countermeasures based on the concept of "constant-encryption-time" were demonstrated against the remote cache timing attack with positive outcomes, in which we establish a secured environment for the AES encryption.