EMITTER International Journal of Engineering Technology

Visual Similarity Detection for Intellectual Property using Deep Transfer Learning

2026-02-19T03:04:38+00:00

Trademarks examination can benefit from deep transfer learning. Utilizing pretrained models to extract image features can significantly improve the trademarks registration process. This approach can facilitate and accelerate image detection. This study aims to enhance the trademark similarity examination process by detecting marks’ visual similarities using deep transfer learning. Deep transfer learning has the potential to develop the registration process of trademarks through the implementation of an automated image detection system, which can enhance detection accuracy. To the best of our knowledge, no automated approach has been used locally to determine the similarities between local trademarks. This study proposes an image similarity detection system to make the trademark examination process more efficient and assist examiners in their decision-making. The proposed system was validated using a dataset provided by the Saudi authority for intellectual property (SAIP). To extract the features, we employed a residual network-based convolutional neural network model (ResNet-50). Then principal component analysis (PCA) was used to reduce the number of extracted features. The proposed system reached a mean average precision (MAP) of 0.774, which indicates a promising result in distinguishing the similarity of trademarks. The findings of this research suggested that an image similarity detection system can support decision-making in trademark examination contexts. Trademark examiners, legal professionals, and intellectual property offices can use the results of this research to enhance their evaluation processes and improve the accuracy and efficiency of trademark registration.

Performance Analysis of Decision Tree Ensemble Models and Feature Importance Analysis in Prediction of Particulate Matter PM10

2026-02-19T03:02:46+00:00

Particulate Matter induced air pollution is known to have significant negative impacts on both the environment and human health. This research evaluates the effectiveness of various decision tree ensemble models in predicting daily PM10 concentrations in Thiruvananthapuram, Kerala, from July 2017 to December 2019. Seven decision tree ensemble models, namely Random Forest, Extra Trees, Gradient Boosting, AdaBoost, LightGBM, XGBoost, and Histogram-Based Gradient Boosting are employed here. To address missing data in the dataset, kNN imputation is utilized for a cohesive dataset suitable for model training. The models utilize both meteorological and air pollutant variables, with performance assessment using metrics such as the coefficient of determination (R²), root mean square error (RMSE) and mean absolute error (MAE). The findings indicate that the Extra Trees regression model provided the best prediction performance (R² = 0.9397, RMSE = 6.664 μg/m³, MAE = 4.950 μg/m³). Histogram-Based Gradient Boosting and Random Forest also demonstrate strong predictive capabilities. The explainability of the best prediction models is conducted by the feature importance analysis process. Feature importance analysis highlighted sulfur dioxide (SO2) as the most significant pollutant influencing PM10 levels, alongside meteorological factors like wind speed and rainfall, enhancing both prediction accuracy and interpretability of results. This research represents the first comprehensive effort to predict PM10 levels in Thiruvananthapuram using machine learning techniques, addressing a gap in regional air quality studies.

FloYO-Net: Enhancing Small Floating Waste Detection in Natural Waters Using Atrous YOLOv5s

2026-02-19T03:00:56+00:00

Detecting small and partially hidden objects in rivers and water bodies remains a major challenge for real-time waste detection systems. These objects are often missed due to their small size, low contrast, and cluttered surroundings. Further complicating the task is the lack of dedicated datasets focused on small floating debris, limiting the development of more capable detection models. To bridge this gap, we developed D_six, a custom dataset of 495 high-resolution images capturing six classes of floating waste under real-world conditions. In this study, we improve the YOLOv5s object detection model by integrating atrous convolutions at three key backbone layers: P1/2, P3/8, and P5/32. These layers represent different scales of the feature pyramid, and the strategic placement of atrous convolution at each level plays a crucial role in helping the model recognize small and occluded objects more effectively. Using a dilation rate of 6, the model’s receptive field is expanded without increasing its size or slowing it down. When trained and evaluated on the D_six data set, the FloYO-Net (Floating Object YOLO Network) consistently outperformed the standard YOLOv5s, achieving a mean Average Precision (mAP@0.5) of 0.828 and mAP@0.5:0.95 of 0.509, compared to 0.787 and 0.498 respectively. Improvements were especially notable for hard-to-detect items like plastic bottles and plastic drink containers, with average precision gains of 6.6% and 7.1%, respectively. These results demonstrate that atrous convolution — when thoughtfully placed — can significantly improve detection accuracy, making it a powerful enhancement for real-time environmental cleanup systems.

Enhanced Wingsuit Flying Search (EWFS) Algorithm for Combinatorial T-way Test Suite Generation

2026-02-19T02:59:05+00:00

The Wingsuit Flying Search (WFS) algorithm is a newly developed global meta-heuristic algorithm. It is efficient and easy to implement, requiring no parameter tuning apart from the population size and the maximum number of iterations. Recently, WFS has been developed based on applying t-way strategies, where t represents the interaction strength. Despite the encouraging results, WFS's search strategy leans more toward local optima due to the narrowing of the boundary search space and the increased value of the search sharpness. Hybridising two or more algorithms enhances search performance by effectively balancing the strengths and mitigating the weaknesses of each method. Thus, this paper proposes a new hybrid Lévy Flight with Wingsuit Flying Search (WFS) algorithm called Enhanced Wingsuit Flying Search Algorithm (EWFS). EWFS uses a control mechanism to identify the best dynamic solution during runtime. The Lévy Flight motion helps the solution escape from local optima and improves the searching process when it gets stuck. Comparison between EWFS and WFS uses the benchmarking configuration of CA(N; 2, 5⁷), while the comparison with other metaheuristic algorithms is based on the following covering array configurations: CA(N; t, 3^p), CA(N; t, v⁷), CA(N; 2, 2^p), and CA(N; t, 2¹⁰). The experimental result shows that EWFS is statistically better regarding test suite size reduction than the recent t-way strategies. It also offers improved results of 65% over the original WFS and resolves the issues of excessive exploitation and getting stuck in local minima or maxima.

Towards Robust Recognition of Handwritten Arabic Characters with Diacritics Using an Incremental Learning Approach Based on CNNs

2026-02-19T02:57:14+00:00

Handwritten Arabic text recognition (HATR) presents unique challenges due to complex character shapes, contextual variations, cursive connections, and the presence of diacritical marks. This study introduces AHAD (Arabic Handwritten Alphabet with Diacritics), a novel benchmark dataset of 71,061 handwritten Arabic character images annotated with five primary vowel diacritics; Fathah, Kasrah, Dammah, Shaddah, and Sukoon, covering 492 distinct classes that combine character identity, contextual form, and diacritic. Leveraging this dataset, we propose an incremental learning framework based on Convolutional Neural Networks (CNNs) to address fine-grained recognition of handwritten Arabic characters with its corresponding diacritics. The model was initially trained on a 114-class dataset of handwritten Arabic characters (in all contextual forms) of non-diacritic characters and fine-tuned in two phases using the AHAD dataset. The two-phase strategy includes output layer expansion, learning rate adjustment, and gradual unfreezing of deeper layers to enhance knowledge retention and prevent catastrophic forgetting. The proposed method achieved a validation accuracy of 92.96% and a test accuracy of 93.26%. Our findings demonstrate the effectiveness of incremental learning for diacritic-aware Arabic handwriting recognition and establish AHAD as a strong baseline for future research in this field.

The Impact of Social Force Model Parameters On Frontier-Based Exploration Performance

2026-03-05T02:32:22+00:00

Autonomous exploration is one of the most challenging tasks in mobile robotics, particularly in environments that contain dynamic obstacles and require fully autonomous mapping without human intervention. This study addresses the dual problem of enabling navigation in the presence of potential static obstacles and achieving autonomous map building. To solve this, we utilize the Social Force Model (SFM), which offers a behavior-based approach suitable for dynamic and uncertain environments. The objective of this research is to investigate how different SFM parameters—Gain (k^s), Radius (r_R), and Effective range (ψ^s)—influence the effectiveness of autonomous exploration. Experiments were conducted using a TurtleBot3 robot in a simulated 155 m² environment, where various parameter combinations were tested. Evaluation metrics included mapping completion, failure types, travel distance, and exploration duration. Results indicate that tuning the SFM parameters significantly affects the robot's ability to explore autonomously and avoid obstacles. Extremely low parameter values led to collisions, while excessively high values caused unstable or inefficient behavior. The Radius parameter had a major impact on spatial awareness, and moderate effective range values contributed to stable tracking. Furthermore, higher frontier sensing latency resulted in longer exploration times. This study provides practical insights into the sensitivity of SFM parameters and offers guidance for optimizing navigation systems for fully autonomous exploration in both simulated and real-world settings.

Machine Learning Approaches for Subcluster in IoT Sensor Networks with Hierarchical Clustering and Dendrograms

2026-03-05T02:32:04+00:00

This research focuses on optimizing IoT Sensor Networks (ISNs) by implementing hierarchical clustering algorithms. Traditional clustering methods often lead to imbalanced energy consumption, impacting network lifetime and performance. Our approach leverages hierarchical clustering to partition the network into a set of clusters. Each cluster has a cluster head and a set of sensor nodes. To enhance data aggregation and energy efficiency, we introduce subclustering within clusters using dendrograms. We assessed performance metrics using simulation, including energy consumption and scalability. The proposed hierarchical clustering methodology significantly improves network lifetime, energy efficiency, and data aggregation.

Sentiment Analysis Design and Development for Low Resource Languages in the Case of Telugu

2026-03-05T02:32:10+00:00

The use of sentiment analysis has become more widespread because it is necessary to filter and analyze information on the internet. It has a wide range of applications, including monitoring social media market research and opinion mining. Still, this development is restricted to few languages with enough resources. The Telugu language lags behind in this field of study, even though it is the fourth most spoken language in India and generates a vast quantity of data every day. In this research paper, we develop a trustworthy source for sentiment analysis in Telugu. To use in sentiment analysis, the data is annotated with Telugu movie reviews. We extracted 1844 sentences from 100 film reviews. We annotated the data with two annotators and calculated the kappa coefficient to determine the annotators' inter-rater reliability. We obtained a kappa value of 0.90 for 1844 sentences, indicating nearly perfect agreement. After the annotators' disagreements and discrepancies were resolved, 1807 sentences were chosen. For feature extraction, we used two vectorization methods: TF-IDF and Count vectorization. Using the two vectorization methods, we used SVM and Logistic regression. We used two vectorization approaches to test different split ratios such as 80-20%, 70-30%, and 60-40% on SVM and Logistic regression. The outcomes of the various combinations are compared. We discovered that combining TF-IDF with SVM for a 70-30% ratio yields the highest accuracy among the combinations tested on our dataset.

Design of Optimized Hardware Architecture for Discrete Cosine Transform using Loeffler’s Algorithm

2026-03-05T02:32:16+00:00

The Discrete Cosine Transform (DCT) is the most commonly used transformation technique in image processing applications for data compression. It transforms a finite set of data points or pixels into frequency domain in terms of sum of cosine coefficients of various frequencies. This paper proposes an optimized hardware architecture for 2D 8x8 Discrete and Inverse discrete cosine transforms (DCT and IDCT) using Loeffler’s algorithm. The hardware architecture is optimized using efficient adders and multipliers. Modified carry select adders (CSLA) are used to boost the speed, wherein Booth multipliers improve overall performance of the design. The Loeffler’s algorithm consisting of 8-stage pipelined architecture reduces the arithmetic operations per cycle and improves processor efficiency. The front-end RTL design of the proposed architecture is implemented on Virtex-7 FPGA, while the backend design is implemented in Cadence using 45nm CMOS technology. The proposed design possesses 24% lesser area, 25% lesser leakage power and 8.8% lesser delay than the existing designs.