https://doi.org/10.5573/IEIESPC.2026.15.1.1

(Shun-yuan Yu)

Finding a truth-value image to use as a reference poses a challenge when assessing the effectiveness of image dehazing. The assessment process is particularly difficult due to the manner in which a 3D model is utilized to synthesize the reference image. Consequently, this paper introduces a novel approach for objectively evaluating the effectiveness of haze removal. By utilizing the inverse degradation of the dehazed image as a guideline, a haze-added image is first created. Subsequently, the difference between the haze-added image and the original hazy image is calculated using the Structural Similarity Index Measure (SSIM). During the inverse degradation process, the atmospheric light is initially searched for within the image’s restricted area using the quad-tree search concept. The transmission is then estimated utilizing the blue channel prior, and the inverse degradation image is ultimately produced by solving the atmospheric scattering model. The results demonstrate that the objective index determined by the proposed algorithm aligns with the observer’s subjective evaluation of the scene.

https://doi.org/10.5573/IEIESPC.2026.15.1.12

(Maaz Ahmed) ; (Kang-Sun Choi)

This paper introduces a novel computational method for the recognition and grouping of characters printed with ancient movable metal types from Korea. Traditional comparative analysis methods struggle to consistently distinguish between typeface variations in digitized images of historical texts. Our approach addresses this limitation through a three-step process: skeletonization to extract structural features, image registration to align, and similarity metric computation to assess typeface matches. We evaluated various pre-processing techniques and metrics, identifying the Dice coefficient as the most reliable for character discrimination. The proposed algorithm combines binarization, skeletonization, iterative closest point registration, and distance measurement. Compared to conventional methods, our approach improves character grouping accuracy while reducing computation time by a factor of up to 45. This enhanced methodology allows for a more accurate reconstruction of the scale and variety of metal types produced, offering new insights into historical printing technologies and cultural developments.

https://doi.org/10.5573/IEIESPC.2026.15.1.25

(Kevin Hung) ; (Gary Man-Tat Man) ; (Jincheng Wang)

The early detection of Alzheimer’s disease (AD) through widespread screening has emerged as a primary strategy to mitigate the significant global impact of AD. EEG measurements offer a promising solution for extensive AD detection. However, the intricate and nonlinear dynamics of multichannel EEG signals pose a considerable challenge for real-time AD diagnosis. This paper introduces a novel algorithm, which is based on Quaternion Principal Component Analysis (QPCA) of multichannel EEG signals, for AD classification. The algorithm extracts high dimensional correlations among different channels to generate features that are maximally representative with minimal information redundancy. This provides a multidimensional and precise measure of brain connectivity in disease assessment. Simulations have been conducted to evaluate the performance and to identify the most critical EEG channels or brain regions for AD classification. The results reveal a significant drop of connectivity measure in the alpha bands. The average AD classification accuracy for all 4-channel combinations reached 95%, while some particular permutations of channels achieved 100% accuracy rate. Furthermore, the temporal lobe emerges as one of the most important regions in AD classification given that the EEG signals are recorded during the presentation of an auditory stimulant. The selection of key parameters of the QPCA algorithm have been evaluated and some recommendations are proposed for further performance enhancement. This paper marks the first application of the QPCA algorithm for AD classification and brain connectivity analysis using multichannel EEG signals.

https://doi.org/10.5573/IEIESPC.2026.15.1.42

(Rong Zhang)

With the continuous development and popularization of Internet technology, online English learning system has become the first choice of many students and educational institutions. However, pronunciation accuracy has always been an important challenge for online English learning systems. The traditional pronunciation classification method has some problems, such as low accuracy and low computational efficiency. To solve this problem, deep confidence network and support vector machine are integrated to construct a pronunciation classification error detection model. The results showed that the loss value of the model tended to be stable after 869 iterations, indicating a good learning effect. The bilingual evaluation BLEU value reached 0.85, indicating a high degree of agreement between the model evaluation and the manual evaluation. The accuracy rate of 91.69% and the F1 value of 0.83 proved the high efficiency of the model in pronunciation classification and error detection. The model achieved 100% feature recognition accuracy in 206 iterations, demonstrating its ability to quickly learn and adapt to pronunciation features. The excellent performance of these performance indicators directly improves the accuracy of pronunciation training in online English learning systems. It also proves that the model is a powerful tool for more effective application of speech recognition and artificial intelligence technology in the field of education, thereby improving learning efficiency.

https://doi.org/10.5573/IEIESPC.2026.15.1.55

(Shunmin Su) ; (Shuangshuang Yan)

In this study, a key gesture recognition method of basketball based on improved YOLOv3 algorithm is proposed. By optimizing the algorithm, by improving the feature extraction network and introducing spatiotemporal information, we construct a data set containing a large number of basketball actions for experimental verification. The enhanced algorithm demonstrated a 92% success rate in recognizing crucial poses, outperforming the original YOLOv3 algorithm by 10%, and shows higher efficiency in real-time video processing, and the average processing time per frame is shortened by 20%. The improved YOLOv3 algorithm shows high performance in basketball’s key gesture recognition tasks, and provides a new technical means for scientific training and competition analysis of basketball. The algorithm combines the spatial domain of convolutional networks to enhance the sensory field of the model and extracts the data of human feature points from two dimensions to improve the deficiency of the model fitting ability. Meanwhile, a multi-head attention mechanism is used to enhance the model adaptive capability. In the experimental tests, the proposed model improvement term has a certain performance improvement compared with the traditional algorithm, and the error index of the model is reduced by 1.15 mm and the accuracy is improved by 2.3% compared with the optimal algorithm, which proves the effectiveness and superiority of the proposed method.

https://doi.org/10.5573/IEIESPC.2026.15.1.67

(Ying Zhang) ; (Xiaoqian Liang) ; (Shenning Yue)

Artificial intelligence and machine learning enhance speech evaluation technology for objective analysis. Strengthening the development of English pronunciation evaluation techniques or tools plays an important role in helping learners correct pronunciation errors and improve their oral proficiency, which cannot be ignored. Traditional evaluation methods have subjective limitations and the selection of data information is one-sided. So this study proposes a recognition model that integrates acoustic phonetics features, taking into account the language habits and pronunciation characteristics of different learners. Time warping and support vector regression are introduced to improve the original Gaussian mixture model-hidden Markov model for speech recognition. And a method is designed for detecting easily confused pronunciation errors, achieving effective fusion of evaluation features from different dimensions. The results confirm that the proposed method achieves a maximum rating accuracy of 90.140% in four aspects of English pronunciation quality. The Pearson correlation coefficient value between it and manual scoring approaches 0.934, which is much higher than the comparison algorithm. Its pronunciation fluency and quality level are both good, and its resource consumption is relatively small. This pronunciation evaluation method can provide technical tools for evaluating students’ oral English proficiency and improving teaching quality.

https://doi.org/10.5573/IEIESPC.2026.15.1.83

(Li Wei) ; (Jingtao Man) ; (Xiufang Wang)

Scenic area roaming is achieved through panoramic roaming technology, which is limited by the quality of image stitching. Therefore, to improve the authenticity of scenic area roaming and promote the development of the tourism industry, a panoramic image stitching technology based on ray vectors and nonlinear fusion algorithms was proposed. The study first used ray vectors to decouple feature points from lens distortion, to solve the problem of feature point selection in fisheye images and ensure the accuracy of image registration. Then, the optimal suture line algorithm was improved through nonlinear weighted fusion algorithm to eliminate the seam seam. The experimental results showed that in the 140? fisheye image, the number of inners for the fast library for approximate nearest neighbors-random sample consensus, channel attention and feature slicing description, and random sample consensus algorithm were 1348, 1763, and 1101, respectively, with inlier rates of 9.6%, 10.3%, and 9.1%. The number of inners and inner rate of the raised algorithm were 2802 and 25.2%, respectively, which are higher than other algorithms. The average root mean square errors of the fast library for approximate nearest neighbors-random sample consensus, channel attention and feature slicing description, and random sample consensus algorithm were 12.2, 11.3, and 17.9, respectively. The average root mean square error of the proposed algorithm was 10.0, which is lower than other algorithms. In addition, the average peak signal-to-noise ratio and structural similarity index of the proposed algorithm were 27.6 and 0.87, respectively, which are higher than other algorithms. The above results indicate that the panoramic image stitching technology proposed in the study can achieve high-quality stitching and natural transition of fisheye images. This method can not only provide tourists with good virtual reality roaming experience and promote the development of tourism, but also has great significance to improve the splicing quality of panoramic splicing and enhance the application value of panoramic image.

https://doi.org/10.5573/IEIESPC.2026.15.1.96

(Lei Li) ; (Yuemei Ren)

Aiming at the image classification problem in small-sample learning, we study the classification algorithm based on the image deformation network, and propose the improvement strategy of modifying the way of selecting auxiliary graphics and modifying the value of image fusion weights. The strategy combines the relational network with the Euclidean distance and includes the RGB channel parameters in the training range as well. On this basis, the dynamic adaptive fusion strategy is further introduced to avoid the target region being segmented too finely. The experimental results show that the image deformation meta-network achieves the best performance in both 1-shot and 5-shot classification tasks, with accuracies of 59.17% and 74.66%, respectively. The improved relational network-deformation meta-network shows improved performance by achieving 59.92% and 75.23% accuracy in 1-shot and 5-shot tasks, respectively. Further the network with dynamic adaptive fusion strategy achieves an accuracy of 60.15% ± 0.28 on the 5-way 1-shot small sample image classification task, which is significantly better than the other strategies. The experimental results show that the improvement of small-sample image classification algorithm based on image deformation meta-network is effective and can significantly improve the classification accuracy, especially the dynamic adaptive fusion strategy, which plays a positive role in improving the classification accuracy

https://doi.org/10.5573/IEIESPC.2026.15.1.108

(Feng Liu) ; (Zongchen Liu)

In the landscape of the Industrial Internet of Things (IIoT), efficient resource allocation remains a pivotal challenge. Addressing this, we introduce a novel distributed algorithm that synergizes with the dynamic nature of IIoT systems. Grounded in the pragmatic constraints of energy, computation, and channel availability, our model innovates through its integration of the Age of Information (AoI) as a central decision-making parameter, ensuring timely data processing and system responsiveness. The algorithm’s distributed nature allows for scalability and adaptability, crucial for the fluctuating demands of industrial settings. Preliminary simulations suggest its potential to optimize resource management, indicating its readiness for further empirical validation.

https://doi.org/10.5573/IEIESPC.2026.15.1.123

(Jianfeng Huang) ; (Qiang Wan)

To address the issue of user emotion and movement recognition in the context of immersive experiences, a multi-source information fusion method based on intelligent wearable devices and brain-computer interface technology is proposed as a potential solution. Its main goal is to further enhance the user’s sense of the immersive experience space, strengthen the collection of the user’s physiological indicators and action information, and then improve the identification accuracy of emotional state and action performance. The experimental results showed that the proposed method could recognize the user’s emotional state with 97.3% accuracy, and the movement recognition with over 90% accuracy. The prediction accuracy of the improved multi-source information fusion technique was 0.98. The results reveal that the method can effectively capture the physiological response of users and realize efficient emotion and movement recognition. It provides a new solution for enhancing user experience, and has important application value for virtual reality, game entertainment, sports rehabilitation and other fields.

https://doi.org/10.5573/IEIESPC.2026.15.1.137

(Duanjiao Li) ; (Gao Liu) ; (Ruchao Liao) ; (Changyu Li) ; (Junsheng Lin) ; (Feng Zhang)

Nowadays, in addition to foreign object damage, there is also network security damage in the form of external force damage to transmission lines. Therefore, traditional methods are no longer suitable for the current external force damage prevention system of transmission lines. In response to such problems, the convolutional block attention module is used to improve the object detection algorithm, and an image detection model is proposed. A new transmission line external force damage prevention system integrating the object detection algorithm and intrusion detection algorithm is proposed by integrating the image detection model with the intrusion detection algorithm. The proposed image detection model achieved the highest classification accuracy of 95.45%, 89.78%, 90.02%, and 96.03% for images of bird nests, balloons, kites, and leaves, respectively. The classification accuracy of the transmission line external force damage prevention system for four network attack methods, including normal recording, denial of service attacks, illegal access from remote machines, and illegal access of local superuser privileges by ordinary users, reached the highest values of 97.36%, 95.23%, 96.31%, and 95.01%, respectively. Therefore, a new model combining two algorithms effectively improves external force damage prevention’s dynamic adaptability and accuracy for transmission lines, which provides an efficient and accurate solution.

https://doi.org/10.5573/IEIESPC.2026.15.1.148

(Se-Young Kim) ; (Sung-Woo Choi) ; (Seong-Won Lee)

In this paper, we present a miniaturized hardware-in-the-loop (HIL) framework designed for the validation of battery management systems (BMS) in electric vehicles (EVs). The framework combines a hardware battery simulator, a motor-driving circuit, and a control computer to replicate realistic driving conditions in a safe and efficient manner. The BMS under test adopts a hybrid algorithm that integrates coulomb counting with open-circuit voltage (OCV) estimation. Validation experiments were carried out using international driving cycles, including US06, UDDS, HWFET, WLTP, EUDC, and J1015. Our experimental analysis demonstrated that the proposed framework, incorporating the Target BMS algorithm, successfully identified cumulative SOC estimation errors under load conditions. Furthermore, this revealed limitations that cannot be fully detected through software simulation alone. These findings suggest that the proposed HIL approach can shorten development time, reduce cost, and improve safety, while also providing a scalable foundation for future BMS research.