Sistema Regional de Información
en línea para Revistas Científicas de América Latina,
el Caribe, España y Portugal

Retaining the effectiveness but improving the efficiency of natural image classification is of prime necessity in recent times, with the surge in demand for deploying these models in practical applications, ensuring accuracy and generalization. Classic deep learning classifiers suffer from limited robustness, generalization, and failure to adapt to new tasks and domains. These shortcomings restrict their practically effective deployment by the availability of different diversified and unseen data. In this work, the authors introduce an optimized deep learning classifier framework, leveraging state-of-the-art techniques in various key domains. The proposed model harnesses a combination of techniques ranging from AugMix, SE-ResNeXt, MAML, Hyperband, and finally Domain-Adversarial Neural Network (DANN) for performance improvement. AugMix integrates Mixup and CutMix with the stochastic augmentation technique of complex augmentation chains to enhance the model's robustness and generalization. Mixing images with stochastic augmentations and the use of Mixup and CutMix bring further strong regularizations, boosting the robustness metrics by 15-20% and classification accuracy by 3-5% on the unseen natural images and samples. SE-ResNeXt introduces the use of channel-wise attention to enhance the representational power of the model. Squeeze-and-Excitation (SE) blocks are introduced to recalibrate the channel-wise feature responses by weighting informative features and suppressing less useful ones. It boosts the accuracy of models on benchmark CIFAR-100 dataset samples by 2-3% over standard ResNeXt. Execution of Model-Agnostic Meta-Learning enables a model to adapt quickly to a new task based on a small number of examples. MAML meta-learns updated models based on examples of tasks instead of direct model parameters. A 5-7% improvement in accuracy is achieved for different scenarios. Hyperband performs tension-free search of optimal hyperparameters via adaptive resources dealing, which configures the resources only for the promising configurations. Reducing the computational cost of hyperparameter tuning to at most 50% ensures an increase in model accuracy of 2- 3%. The DANN technique uses adversarial training in order to suppress the domain shift between source and target datasets. DANN uses a gradient reversal layer to train feature extractors to produce domain-invariant features, leading to a 10~15% increase in accuracy on target domain datasets compared to non-adaptive methods.

Polycystic ovary syndrome (PCOS) disorder is caused by a protracted menstruation cycle that frequently elevated the androgen levels of women in their reproductive age. Insulin resistance affects 50% to 70% of all women with PCOS, and hormone difference contributes the high levels of testosterone that causes the symptoms and signs of PCOS. This work develops a deep learning (DL)-based PCOS diagnosis to address these issues. At the initial stage, the ultra sound image is preprocessed by means of adaptive Wiener filter for noise removal process. The Polycystic ovary (PCO) follicles segmentation process is performed using the Fuzz Local C-Means Clustering (FLICM). Feature extracti­­­on is the neat stage, where the Speeded-Up Robust Feature (SURF), Shape index histogram as well as the statistical features includes variance, mean, kurtosis, entropy and standard deviation are extracted. Furthermore, the PCOS detection is done in the next stage, where a deep Q Net (DQN) is utilized and the parameters of DQN is optimized by the adaptive Archimedes optimization algorithm (AOA). Moreover, the system performance is evaluated using accuracy, sensitivity and specificity parameters with the corresponding values like 0.906, 0.918 and 0.928.

PWR is a type of nuclear reactor that is widely used as a nuclear power plant. Even though PWR has been around for a long time,  technology continues to develop. The direction of development of PWR technology is to create a more compact design with a modular system (SMR) and more efficient fuel. More efficient fuel can be obtained by increasing fuel burnup. By increasing burnup, the fuel usage period is longer, thereby increasing the economic value of the fuel and reducing the volume of radioactive waste produced from spent fuel. High burnup means the fuel will be exposed to radiation for longer. Therefore, it is necessary to calculate both thermal and mechanical aspects with the new fuel rod design, to see whether the fuel can be used until the end of the fuel cycle. Calculations wereconducted using the Femaxi version 6 code. From the calculation results, it was obtained that the dimensions of the fuel rods were capable of reaching a burnup of 60 GWd/TU. The dimensions obtained include the diameter and length of the pellets of 7.4 mm and 10 mm, the diameter and depth of the disc of 4.7 mm and 0.51 mm, and the inner and outer diameters of the cladding of 7.8 mm and 9.3 mm. The calculation results show that the temperature distribution in the fuel rods during reactor operation is still within safe limits, and pellet cladding interaction (PCI) does not occur until the end of the fuel consumption cycle.

In this research, we used a combination of experimental techniques to shed some light on the effect of lithium perchlorate (LiClO(4)) on the electrochemical and physical properties of indium tin oxide (ITO) films. For this, we studied the effect of ITO films immersed in a LiClO(4)/PC+EC solution.  Chronoamperometry, along with transmissivity measurements, shows that ITO undergoes optical changes as a function of voltage changes. However, as the number of cycles increases, the transmissivity decreases significantly due to ITO degradation. Cyclic voltammetry reveals the presence of reduction and oxidation processes, suggesting ITO reduction and possible ion trapping. X-ray  diffraction does not show compelling signs of lithium insertion or the presence of other phases. Finally, X-ray photoelectron spectroscopy is used to evaluate the oxidation states and chemical bonding. The analysis reveals the presence of indium oxide, suggesting that the surface is mainly ITO, in agreement with XRD studies.

In this work, a thermal and acoustic study of specimens made from coconut fiber agglomerated with cassava starch is carried out. Sound absorption was measured in a transmission tube according to ISO 10534-2: 2001, in order to obtain the sound absorption coefficient (α). In addition, procedures described in ASTM E2611–19 were implemented to determine the sound transmission loss (STL). The results demonstrate      the capacity of the tested specimen as a sound absorber, with absorption coefficients greater than 70% for a considerable range of frequencies starting at around 1000 Hz and above. Similarly, the thermal study of the material based on ASTM C-177 indicates an average thermal conductivity coefficient of 0,174 W/m.K, in a range of inlet temperatures between 52°C and 137°C, confirming that it has qualities that are similar to good thermal insulators, although still not comparable to some industrial materials.    

In the realm of optical fiber communication systems, maximizing transmission efficiency stands as a paramount objective. This study embarks on an innovative approach, merging wavelengthdivision multiplexing (WDM) with dispersion compensation fiber (DCF), to address the persistent challenges of signal degradation due to dispersion. Drawing from comprehensive simulations and meticulous analysis, our research reveals the transformative potential of this integrated solution. By seamlessly integrating WDM and DCF, we achieve remarkable enhancements in transmission performance, characterized by superior signal fidelity, unprecedented transmission distances, and unparalleled data rates. This study not only underscores the technological advancements propelling optical communication systems into a new era of efficiency but also heralds the dawn of a paradigm shift in high-speed, long-distance communication networks.

This paper presents the kinematic and dynamic modeling of the shoulder-clavicle assembly constituting a four–degrees-of-freedom mechanical system. The models are obtained through robotic concepts and formulations, applied to a specific case of arm abduction with movement in the acceleration and deceleration phase, and compared with its equivalent static model. The influence of a suspended mass at the arm´s end is also analyzed. Subsequently, a biomechanical model considering the muscular action of the deltoid muscle is created based on the dynamic model obtained, allowing estimation of the force exerted by the moving muscle.

Wireless communication is being modernized with Reconfigurable Intelligent Surface (RIS) antennas, which enhance signal coverage, capacity, and energy efficiency. This project depicts the Rogers RT duroid 5880 substrate design and simulation for a 28 GHz RIS antenna. Phase stability with varactor integration was without warning detected in initial probations using a 'P' shaped unit cell. The design was polished to a 'R' shape and a stripline was added to boost phase response and delegate for dynamic phase modification. For electromagnetic simulations, CST software was utilized; however, MATLAB conceivable accurate phase visualization, hence rout CST's downsides. The resulting design shows amplified system efficiency, beamforming, and variability. However, because of analytical limitations, simulating a 32 by 32 RIS array caused difficulties. In the face of this, the discoveries highlight how RIS antennas can luxuriously improve wireless communication performance, especially in knotted settings.        

The underwater acoustic sensor network is a large network consisting of many operating sensor nodes that surround a transmitting node. The communication process faces substantial disturbances caused by the everchanging nature of the underwater acoustic channel, which is characterized by fluctuating properties in both time and location. Therefore, the underwater acoustic communication system has difficulties in reducing interference and improving communication efficiency and quality by using adaptive modulation. This work presents a model that aims to tackle these difficulties by suggesting an optimum route selection and safe data transmissionapproach in UASN using sophisticated technology. The suggested approach for transferring safe data in UASN via optimum route selection consists of two main stages. Nodes are first chosen based on restrictions such as energy, distance, and connection quality, which are quantified in terms of throughput. Moreover, the process of forecasting energy is made easier by using sophisticated machine learning methods like transformer models. The ideal route is generated using a hybrid optimization technique called enhanced swarm optimization, which combines ideas from particle swarm optimization and genetic algorithms. Afterward, data is safely transported via the most efficient route by using fully homomorphic encryption. Finally, the ESO+ transformer model that was created is tested against established benchmark models, showcasing its strong and reliable performance. The proposed model demonstrates remarkable performance with an accuracy of 95.12%, precision of 94.83%, specificity of 93.65%, sensitivity of 95.28%, false positive rate of 4.72%, F1 score of 94.95%, Matthews correlation coefficient of 94.85%, false negative rate of 4.72%, negative predictive value of 95.15%, and false discovery rate of 5.15% when trained on a learning percentage of 70%.

Incorporating large-scale multiple-input multiple-output (MIMO) systems in densely deployed renewable energy systems (RES) represents a significant challenge in developing next-generation wireless networks. This field combines cutting-edge communication technologies with sustainable energy systems to enhance network  communication and energy management in smart grid applications. Furthermore, varying energy availability in RES-based environments and dynamic load profiles make it difficult to achieve optimal beam attachment in mmWave massive MIMO systems. Conventional beam attachment techniques perform poorly in such dynamic conditions, resulting in poor network performance and high latency. This has created the need for better and more versatile  approaches to beam attachment that can address this inherent variability of RES while at the same time providing highly accurate and low-complexity solutions. This paper presents an improved beam attachment recognition system explicitly designed to operate in RES conditions. Thus, the innovative strategy presented in this work is based on ensemble learning, which includes Random Forest (RF) and Extreme Gradient Boosting (XGBoost) classifiers, making the prediction more accurate and the system more stable. The proposed method integrates RES-specific signal strength, interference, traffic load, and renewable energy availability into the choice of the preferred beam. Cohesive simulations support our approach in this case. The Random Forest (RF) classifier test accuracy was 97.56%, and the XGBoost classifier was 97.84% – both of which are higher than conventional methods. Analyzing the feature importance of the problem, it was found that distance, angle, and signal strength were the most significant factors in beam assignment. The performance of the system was also very impressive in terms of scalability, with accuracy rates barely flinching even as the number of samples reached 50,000. Also, the energy efficiency analysis showed that the proposed beam attachment approach could lead to more energy-efficient network operations.