Yun Cheng

This paper proposes DCGAN-YOLOv8n, an integrated framework that significantly advances small-sample target detection by synergizing generative adversarial feature enhancement with multi-scale representation learning. The model’s core contribution lies in its novel adversarial feature enhancement module (AFEM), which leverages conditional generative adversarial networks to reconstruct discriminative multi-scale features while effectively mitigating mode collapse. Furthermore, the architecture incorporates a deformable multi-scale feature pyramid that dynamically fuses generated high-resolution features with hierarchical semantic representations through an attention mechanism. The proposed triple marginal constraint optimization jointly enhances intra-class compactness and inter-class separation, thereby structuring a highly discriminative feature space. Extensive experiments on the NWPU VHR-10 dataset demonstrate state-of-the-art performance, with the model achieving an mAP50 of 90.46% and an mAP50-95 of 57.06%, representing significant improvements of 4.52% and 4.08% over the baseline YOLOv8n, respectively. These results validate the framework’s effectiveness in addressing critical challenges of feature representation scarcity and cross-scale adaptation in data-limited scenarios.

With the advancement in deep-space exploration, the injection technology using xenon as a working fluid in electric propulsion systems has emerged as a key area of interest. To delve into the gas-liquid dynamics of the liquid xenon injection process and the influence of gravity on this mechanism, this investigation employs a VOF two-phase flow model coupled with the Lee model to elucidate the characteristics of the two-phase flow during microgravity conditions. The findings uncover that in the absence of gravitational forces, gas-liquid stratification does not occur during the filling process. Consequently, this leads to an even distribution of gas and liquid within the tank, which in turn prolongs the filling duration in orbiting scenarios.

To investigate the phenomenon of liquid xenon flashing in a filling pipeline, the two-phase flow in a pipe is calculated and analyzed by using a one-dimensional homogeneous equilibrium model (HEM) and a two-dimensional mixture model. The distribution of xenon two-phase flow parameters along the pipeline is observed by the numerical solution of a one-dimensional HEM and simulation by Fluent. The comparison and analysis of the results of different models show that the one-dimensional HEM can quickly attach the critical mass flux faster than Fluent’s simulation under the given filling conditions, which verifies the rationality and rapidity of the numerical solution in calculating the flash process. The influence of the diameter and length of the pipeline on the flashing process of liquid xenon is analyzed by a one-dimensional theoretical model. The results show that the geometric parameters of the pipeline have a great impact on the mass flow rate and the position of the initial phase transition point, but have little effect on the void fraction at the outlet. An increase in pipe diameter and pipeline length delays the onset of phase transition. Compared with liquid oxygen and liquid nitrogen, liquid xenon is more likely to undergo a phase transition. The phase change kinetics of oxygen and nitrogen are roughly 70% as fast as those of xenon.

With the continuous development of information technology, network security risks are also rising, and the ability to quickly perceive network threats has become an important prerequisite and an important means to cope with network risks. Currently, there are various types of network attacks and complex attacking techniques, and the large differences between them have led to the difficulty of collecting and recognizing the common characteristics of network attacks. Considering the regular temporal fluctuations in network attacks, a method for network security situational awareness, based on an enhanced Particle Swarm Optimization Bidirectional Long Short-Term Memory (BiLSTM) network model, is proposed. By gathering and organizing critical information within the network, an encapsulated Wrapper feature selection algorithm is utilized for the extraction of element features. The refined Particle Swarm Optimization algorithm is applied to optimize the parameters of the BiLSTM network, enabling a rapid convergence and enhancing the training efficiency, thus effectively identifying the categories of network attacks. The experimental results show that the MAPE for the proposed model has been diminished to 0.36%, while the sMAPE is 2.654%. Additionally, the fitting coefficient attains a value of 0.92. This indicates a high level of recognition and precision exhibited by the proposed model in detecting network security risk behaviors. Furthermore, in contrast to the traditional CNN neural network, the proposed model is more compact, which significantly reduces the computational overhead and allows for efficient network security situational awareness.