Since feature point matching and optical flow estimation are closely related to the image texture, a standalone visual approach for video stabilization may not be suitable for all scenarios was used. A tightly coupled attitude-sensor-based video stabilization method was proposed. By adaptively adjusting weights, the optimal homography between images relies more on the attitude sensor in low-texture areas or more on feature matching in rich-texture areas. After obtaining the optimal transformation, a robust elastic warping method was applied to further align consecutive image frames. Experimental results demonstrate that the proposed video stabilization method achieves better performance and robustness.

Revealing the effect of particle size distribution (PSD) to the stiffness characteristics of the red stratum soil-rock mixture (RS S-RM) has significant implications for the subgrade construction, with Sichuan Basin as a representative. 20 large-scale triaxial tests with different PSDs were conducted. Introducing the concept of effective dominant skeleton size ( d_{\mathrm{e}\mathrm{d}}), digital image processing technology was utilized to collect the d_{\mathrm{e}\mathrm{d}}. The relative dominance ratio ( D_{\mathrm{d}}) describing the structural skeleton changes after the test was defined. The response of the stiffness characteristics of the RS S-RM regarding confining pressure ( {\sigma }_3) and D_{\mathrm{d}} were investigated. Validation experiment and related research verify the credibility of the formation mechanism of stiffness. The results indicate that as the shearing process, more and more particles near the shear plane transition from their initial interlocking state to sliding friction. Consequently, the soil-stone framework evolves from a suspended-dense structure to a skeleton-dense structure, and finally transforms into a skeleton-void structure; the initial deformation modulus ( E_0) relationship with {\sigma }_3 and D_{\mathrm{d}} follows: E_0 = 14.50 + ( 1 + 36.33 {{\sigma }_3}^{0.1} ) {\mathrm{e}}^{1.06D_{\mathrm{d}}}. The relationship between the tangent deformation modulus ( E_{\mathrm{t}}) of the specimen at failure {\epsilon }_1 = 15 \mathrm{\%} is E_{\mathrm{t}} = - 6.06 + ( 1 + 0.08 {\sigma }_3 ) {\mathrm{e}}^{-0.09D_{\mathrm{d}}}. Validation experiment and related research confirmed the response of E_0 and E_{\mathrm{t}} to {\sigma }_3 and D_{\mathrm{d}}. The stiffness characteristics of the RS S-RM can be predicted solely based on the PSD and {\sigma }_3.

Bedding slope is an easy-to-slide structure often encountered in mountain highway construction, and weak interlayer is a typical potential sliding surface. Taking the bedding slope of expansion project of Guangyuan-Mianyang section of G5 Beijing-Kunming highway as an example, the physical and mechanical parameters of the weak interlayer and the water content and strength parameters under different saturation time were obtained through laboratory tests. The continuous-discontinuous numerical simulation was carried out by using the finite element-smooth particle dynamics method to study the instability evolution and failure mode of the bedding slope. The results show that with the saturation softening of the weak interlayer, the water content increases and the strength decreases, and the plastic zone of the bedding slope continues to extend to the trailing edge of the slope. When the water content is higher than the liquid limit of the weak interlayer, the plastic zone range increases sharply to the first instability length, and the slope shows traction slip-crack failure. The horizontal displacement of the bedding slope has a displacement peak point at the foot of the slope at the slope line slope position, and there is a maximum displacement at the first-level slope platform ; with the increase of the water content of the weak interlayer, the horizontal displacement of the slope increases continuously, and the maximum point of the horizontal displacement appears after 1 h of saturation and begins to undergo large-scale instability deformation.

Icing on aircraft poses a serious threat to flight safety, and electric heating is an efficient method for anti-icing and de-icing. A three-dimensional mathematical model that considers water film flow and heat transfer for electric heating anti-icing was constructed. Additionally, a numerical calculation method for electric heating anti-icing was proposed. This method was applied to simulate the steady-state anti-icing process of the NACA0012 airfoil under continuous electric heating conditions. The accuracy of this calculation method was validated by comparison with existing experimental data and computational results. The results indicate that when the heating power is low, the water film flows out of the heated area and overflow ice forms downstream. Under the same inflow conditions, a higher heating power results in a higher anti-icing surface temperature and a smaller water film coverage area. Furthermore, the calculated anti-icing surface temperatures are within 5 ℃ error compared to experimental data, and for inflow temperatures no lower than 6.67 ℃, the error is less than 3 ℃.

The innovative development of core technologies in artificial intelligence field has become a competitive focus for major economies to seize the first-mover advantage. The analysis of the competitive landscape of artificial intelligence technology from the perspective of technical intelligence plays an important role to accelerate related technology innovation layout and decision-making of responding to challenges.With the help of analysis tools like Histcite, VOSviewer, and the “Science Headlines” big data collection tool of the Beijing Academy of Science and Technology, the development trend of artificial intelligence technology was identified based on data from the Web of Science database, global patent database. Combined with the analysis of the United States’ restrictive measures on China’s science and technology, the international competition situation, opportunities and challenges facing the development of China’s artificial intelligence technology were analyzed. The countermeasures and suggestions for the development of China’s artificial intelligence technology are proposed, which can provide reference for promoting the rapid iterative development of AI technology in China

To study the deformation patterns of wet-sinking loess widely distributed in the alluvial plains of the northern Tianshan Mountains in Xinjiang, experimental studies were carried out to differentiate the effects of soluble salts on the wet-sinking and dissolving effects of loess. Through the ‘three-line method’ compression test of original water content saturated with Na₂SO₄ salt solution and pure water on the original loess samples with five kinds of soluble salt contents, the influence of soluble salt content and its state of existence on the characteristics of loess wet subsidence and subsidence was obtained, and the calculation model of the amount of loess wet subsidence and subsidence deformation with different salt contents was constructed. The results show that there is a critical value for the salt content change to improve the skeleton effect between particles, and wet subsidence is the main mode of deformation when it is less than 8‰ and more than 23‰, while dissolution subsidence is the main mode of deformation when it is between 8‰~23‰; mathematical relationship between each salt content and deformation coefficients under the step-by-step pressure has been fitted by using the Giddings and Extreme model function. The FreundlichEXT regression model was used to construct the starting pressure function equation for different salt contents of soil samples in the study area. The research results can provide scientific basis for the study of the mechanism of joint action of wet subsidence and dissolution subsidence of eolian salt in the region, the design of structural loads and the calculation of deformation.

In response to the problem of insufficient accuracy of the traditional Sadovski model, based on vibration monitoring data from the soft rock tunnel blasting construction site of the West Chongqing high-speed railway, the variation trend of the model parameters of the traditional Sadovski model at different footage was explored, and the Sadovski model was improved accordingly. The fitting accuracy of the traditional model and the improved model was compared. The results show that the peak vibration velocity during blasting construction generally decreases exponentially with the increase of blasting center distance, and the traditional model has a large degree of data dispersion after fitting. In the traditional Sadovski model, the parameters k(coefficient of association)and α(attenuation index)exhibit linear and exponential functional relationships with the footage, respectively. The prediction accuracy of the modified model is improved by about 23% compared to the traditional model.

In response to the demand for reservoir protection during well maintenance of low-pressure gas wells in the South China Sea L gas field, the optimal evaluation of the system types and concentrations of foaming agents and foam stabilizers in the laboratory was carried out. A set of foam workover fluid system and preparation process suitable for L gas field were constructed. The system has a half-life of 60~72 h under reservoir conditions of 80~90 ℃ and a maximum temperature resistance of 100 ℃, demonstrating good temperature resistance and stability. The foam density can be as low as 0.5 g/cm³, matching with the pressure coefficient of the target reservoir, and effectively reducing the leakage. Through core damage test, the recovery rate of permeability of high permeability core after being polluted by foam workover fluid can reach 95.1%, indicating that the system has excellent reservoir protection performance.

The state of health(SOH) and remaining useful life(RUL) of a battery are core indicators for evaluating battery performance degradation and potential lifespan. Accurately predicting the SOH and RUL of batteries is crucial in practical applications. To capture changes in battery performance and make predictions, operational data of the battery is typically relied on to train machine learning algorithms, such as neural networks or deep learning methods. However, traditional machine learning models often adopt a single architecture to adapt to the entire dataset, which is insufficient when dealing with complex and highly heterogeneous big data. Such models generally have the risk of insufficient generalization ability and overfitting, and are inefficient in big data processing. Therefore, Hierarchical sparse mixture of experts(HS-MoE) and multi head mixture of experts(MH-MoE) models were used to construct predictive models for battery State of Health(SOH) and Remaining Useful Life(RUL), respectively. Comparative experiments were conducted on publicly available datasets from NASA and EIS, and the results showed that the MH-MoE model outperformed the HS-MoE model in predicting SOH and RUL on both datasets.

Thermoplastic resin is the main film-forming material of hot-melt marking coating, and it can guarantee the mechanical properties and service life of hot-melt marking. By changing the type and content of the resin, the influence of the resin on the mechanical properties of the hot-melt marking coating series was studied. The results show that the resin type has little influence on compressive property and wear property but has a significant impact on adhesion property, shear property, and adhesion property. At the same time, with the increase of resin content, the compressive strength increased first and then decreased, and the wear property, adhesion property, shear property, and adhesion property showed an overall increase trend. When the resin type is C5 petroleum resin and the content is not less than 20%, the hot-melt marking line has excellent mechanical properties.