Traditional object detection methods, such as anchor-based YOLO variants, face challenges due to the irregular shapes and small sizes of these contaminants. This paper introduces a novel approach... Show More

Traditional object detection methods, such as anchor-based YOLO variants, face challenges due to the irregular shapes and small sizes of these contaminants. This paper introduces a novel approach that leverages swarm Intelligence to enhance the performance of a keypoint-driven YOLO framework. By integrating keypoint detection with Boundary-Aware Vectors (BBAVectors) and utilizing swarm intelligence algorithms for model optimization, our approach improves the localization and identification of small, irregularly shaped non-metallic objects. By optimizing the feature extraction process through swarm-based techniques and incorporating keypoint-driven object detection, our model significantly boosts precision and recall compared to traditional methods. Evaluated on a custom dataset of fiber like materials, our approach achieves a mean Average Precision (mAP) of 92.9% at IoU 0.5, demonstrating strong performance in real-world applications.

This paper proposes an advanced framework that combines enhanced particle swarm optimization with least squares support vector machine learning (PSO-LSSVM) for intelligent lightning risk assessment... Show More

This paper proposes an advanced framework that combines enhanced particle swarm optimization with least squares support vector machine learning (PSO-LSSVM) for intelligent lightning risk assessment in overhead power line systems. Our approach integrates comprehensive feature extraction methodologies capturing both temporal and spectral characteristics of lightning phenomena with an optimized classification system for rapid risk evaluation. The PSO algorithm is specifically adapted to determine optimal LSSVM parameters, improving classification accuracy and computational efficiency. Experimental validation using data from three climatically distinct regions - a 220kV tropical coastal line, a 345kV alpine mountain network, and a 400kV desert transmission system - demonstrates the framework's versatility and effectiveness. Results show that our PSO-LSSVM framework achieves 94.6%, 94.3%, and 94.9% classification accuracy in tropical, alpine, and desert regions, respectively, representing substantial improvements over baseline methods.

This paper proposes ENNOEIGS, an evolutionary neural network-based online ecological industrial governance system that integrates advanced neural architectures with evolutionary optimization for... Show More

This paper proposes ENNOEIGS, an evolutionary neural network-based online ecological industrial governance system that integrates advanced neural architectures with evolutionary optimization for robust pollution monitoring. The framework combines convolutional neural networks for dimensional reduction of sensor data, external attention mechanisms for discovering pollution pattern correlations, and convolutional long short-term memory networks for modeling the spatiotemporal evolution of contaminants. A genetic algorithm continuously optimizes the neural network parameters, enabling adaptation to changing industrial conditions. Experimental validation using industrial wastewater monitoring data demonstrates ENNOEIGS's superior performance, achieving a 94.8% anomaly detection rate with 2.3% false alarms, outperforming existing approaches. The framework reduces the mean modified absolute error to 0.028 mg/L while maintaining faster convergence during training.

Due to the influence of camera Angle, camera exposure parameters and other reasons, the color and brightness of art painting images are significantly different. Therefore, this paper proposes a... Show More

Due to the influence of camera Angle, camera exposure parameters and other reasons, the color and brightness of art painting images are significantly different. Therefore, this paper proposes a color correction method for art painting images based on perceptual network for the application of sequence data in wireless systems. When the damaged area is large or the semantic information is missing, the restoration effect will be greatly reduced. Therefore, a high-resolution generative network is designed as the back-end of the network to improve the resolution of the density map and further improve the model counting accuracy. Then, the multi-scale feature extraction network based on the perception network and the high-resolution density map generation network are connected and fused to build the dense crowd counting network based on the multi-scale perception network. The convolutional network using neural art images greatly reduces the feature quantity in the grid extraction of feature combination. This ensures the accuracy of multi-scale artistic images.

This paper proposes an English Translation Model Parameter Adaptive Tuning Method Integrating Particle Swarm Optimization (PSO) and Attention Mechanism. We encode key Neural Machine Translation... Show More

This paper proposes an English Translation Model Parameter Adaptive Tuning Method Integrating Particle Swarm Optimization (PSO) and Attention Mechanism. We encode key Neural Machine Translation (NMT) hyperparameters as particles and employ a specialized fitness function that rewards both conventional translation metrics and attention alignment quality. Furthermore, a dynamic feedback multiplier modulates PSO velocity updates in real time, incentivizing hyperparameter adjustments when attention scores deviate from predefined thresholds. Experimental evaluations on both IWSLT En–De and WMT En–De datasets demonstrate the effectiveness of this method, yielding improvements in BLEU and ROUGE-L scores relative to standard baselines. Ablation studies further confirm that both the attention alignment term in the fitness function and the dynamic attention feedback are indispensable for achieving superior translation quality and more coherent attention patterns.

This paper proposes a novel framework that integrates swarm intelligence algorithms with controllable Generative Adversarial Networks (GANs) to meet the multi-objective demands of advertising... Show More

This paper proposes a novel framework that integrates swarm intelligence algorithms with controllable Generative Adversarial Networks (GANs) to meet the multi-objective demands of advertising illustration tasks. By combining Particle Swarm Optimization (PSO) for global parameter exploration with a StyleGAN-like architecture capable of fine-grained style manipulation, our method effectively balances visual fidelity, brand consistency, stylistic diversity, and computational efficiency. We formulate the generation process as a multi-objective optimization problem. Experiments conducted on a curated dataset show that the proposed method outperforms conventional GANs, conditional GANs, and evolutionary-based baselines. Ablation studies further demonstrate the importance of integrating both style and brand-related loss functions, while parameter sensitivity analyses highlight the role of swarm size, inertia weight, and acceleration coefficients in guiding the search toward optimal solutions.

The recently proposed swarm intelligence Artificial Rabbits Optimization (ARO) performs well, but there are still some drawbacks, including low population diversity, unbalanced exploration and... Show More

The recently proposed swarm intelligence Artificial Rabbits Optimization (ARO) performs well, but there are still some drawbacks, including low population diversity, unbalanced exploration and exploitation capabilities, and low convergence accuracy. To address the above issues, this article proposes a variant of ARO named MARO, which adopts three strategies to overcome the limitations of ARO and improve its performance. This paper uses 23 classic test functions and CEC2017 test functions for testing. The experimental results show that MARO has higher convergence speed, accuracy, and stability than the comparison algorithms. In addition, the enormous potential of MARO in practical applications is further verified through five real-world engineering application problems.