To address the issues of high noise levels, low contrast, and insufficient clarity in pellet CT images—particularly given the demand for high-quality images during 3D reconstruction a pellet image resolution enhancement model based on MSEEGAN(Multi-Scale Edge Enhanced GAN) is proposed. Current image enhancement methods still face limitations in denoising, edge clarity improvement, and adaptability to industrial applications. Traditional approaches struggle to effectively remove complex noise, while deep learning-based techniques may result in the loss of texture information and involve high computational costs, making practical deployment challenging. This study adopts an end-to-end training approach to enhance the discriminability of material regions. During the generation process, edge enhancement loss and pixel attention are introduced to optimise boundary clarity, while multi-scale discrimination and edge-aware loss are incorporated to improve detail representation. Experimental results show that MSEEGAN significantly outperforms other methods in metrics such as AG (87.5), IE (7.01), AEC (7,103,909), and EPR (0.178), enhancing image edge clarity and contrast, and providing higher-quality data for 3D reconstruction. Moreover, the integration of the Industrial Internet of Things (IIoT) framework spans the entire process—from pellet production and computed tomography (CT) scanning to image enhancement and 3D reconstruction—facilitating end-to-end digital management and advancing the intelligent and efficient development of pellet production.

Real-time object detection plays a critical role in advancing autonomous driving technologies. To meet the demands for real-time performance, a lightweight model is essential to reduce parameter size, simplify complexity, and enhance detection speed. This paper presents YOLO-FSE, a compact vehicle detection model built upon the YOLOv5 framework. In YOLO-FSE, the C3 module is substituted with the C3faster module from the FasterNet lightweight architecture. The backbone network is augmented with the Shuffle Attention module, which improves feature fusion by separating spatial and channel attention mechanisms. Additionally, the original loss function is replaced by EIoU loss, which directly penalizes width and height predictions, thereby improving the model’s generalization ability. During experimentation, it was assumed that the number of test images was sufficiently large and clear, with PyTorch serving as the primary framework for model development and training. Experimental results show that, on the UA-DETRAC and BIT-Vehicle datasets, the mean average precision (mAP) achieved 98.9% and 97.6%, respectively. Computational complexity (FLOPs) was reduced by 12.65%, while the average frame rate (Fps) increased by 46.13%. Overall, the enhanced model exhibits substantial improvements in both speed and performance, demonstrating its potential for deployment in autonomous driving applications.

With the continuous development of autonomous driving, real-time target detection has become increasingly critical in autonomous driving systems. However, traditional target detection algorithms usually require huge computational resources, limiting their application on embedded autonomous driving platforms. To address this challenge, a lightweight single-stage target detection algorithm is proposed YOLOdrive. The inverted residual structure, linear bottleneck layer, and depth-separable convolution in MobileNetv2 are utilized to improve the YOLOv8 backbone network, while the spatial channel reconstructed convolution is used to improve the C2f module of YOLOv8, and the convolution of YOLOv8 neck and detection head is replaced by the depth-separable convolution. Experimental results verify that the average accuracy of YOLOdrive algorithm on MS COCO2017 data set and VOC2007 data set is improved compared with the baseline model YOLOv8-Nano. The amount of model parameters has been reduced by more than 50%, and the amount of computation in the model has been reduced by more than 70%. The algorithm drastically reduces the amount of parameters and computational complexity of the network, improves the operational efficiency, saves the storage space of the network, and maintains a high-detection performance.

The current unreasonable deployment mechanism of shared bikes makes it difficult to meet public demand. Accurate prediction of shared bikes demand can alleviate this problem, but current prediction models are limited by small data size and poor robustness, making them unable to provide sufficiently accurate prediction to be applied in real-life situations. In this work, we propose a generative adversarial networks (GAN)-based ensemble model (GANEM) for predicting the demand of shared bikes in 5G networks to solve the above problem. With the support of 5G networks, real-time transmission and storage of traffic data of shared bikes have become reality. We then use these data to train the GANEM model to have better generalization performance. The GANEM model first uses GAN to learn the data features of existing shared bikes demand to generate new training samples. We borrow the idea of ensemble learning, three benchmark predictors are integrated to fully exploit the association between explicit features, which can improve the robustness of the model. Excellent prediction results are achieved on two real datasets. The experimental results of our proposed GANEM model on two datasets are as follows, the values of MAE are 0.120 and 0.144, the values of MSE are 0.027 and 0.032, the values of RMSE are 0.164 and 0.181.

A rational approach to predict passenger car sales and analyze the current state of the passenger car sales market can contribute to the healthy development of the automotive industry. The number of features used to describe passenger car sales in real life is too redundant. The features of human empirical filtering and combination cause loss of time. In addition, explicit features are too homogeneous, which need to be complemented by implicit features. The fixed-parameter weights in the trained model cannot provide some estimates for unknown uncertainties. Therefore, in this work, we propose a DeepFM-based nonparametric model (DFMNP) for predicting passenger car sales in sustainable way. We use a big data platform to provide data for the DFMNP model. The DFMNP model uses feature engineering to expand the number of explicit features, a multilayer neural network to extract implicit features, and a Bayesian neural network to replace the neural network with fixed weights for inferring predictive values. In addition, a factorization machine is used in the prediction function to take into account the cross information of implicit features. The combination of the above improvement points can be used to improve the model’s expressive and predictive power for unknown data. Its prediction performance on two real passenger car sales datasets is as follows, RMSE values of 0.0825 and 0.116, and MAE values of 0.0482 and 0.0595. The above experimental results verify the superiority of the method proposed in this work.