Small object detection in traffic sign applications often faces challenges like complex backgrounds, blurry samples, and multi-scale variations. Existing solutions tend to complicate the algorithms. In this study, we designed an efficient and simple algorithm network called StarCAN-PFD, based on the single-stage YOLOv8 framework, to accurately recognize small objects in complex scenarios. We proposed the StarCAN feature extraction network, which was enhanced with the Context Anchor Attention (CAA). We designed the Pyramid Focus and Diffusion Network (PFDNet) to address multi-scale information loss and developed the Detail-Enhanced Conv Shared Detect (DESDetect) module to improve the recognition of complex samples while keeping the network lightweight. Experiments on the CCTSDB dataset validated the effectiveness of each module. Compared to YOLOv8, our algorithm improved mAP@0.5 by 4%, reduced the model size to less than half, and demonstrated better performance on different traffic sign datasets. It excels at detecting small traffic sign targets in complex scenes, including challenging samples such as blurry, low-light night, occluded, and overexposed conditions, showcasing strong generalization ability.

The relationship between natural resource development and sustainable development has long been a focus in academia. In the context of a new global economic cooperation system, many scholars argue that such cooperation can lead to a “resource curse” effect in partner countries, hindering their sustainable development. This study analyzed panel data from 64 countries from 2008 to 2020, using the Belt and Road Initiative as a representative of multilateral economic cooperation (MEC) policies. The aim was to examine the actual impact of multilateral economic cooperation on the sustainable development levels of partner countries and to explore the underlying mechanisms influencing these outcomes. First, we measured and identified the sustainable development index (SDI) under natural resource development schemes and the “resource curse” effect in these countries. Then, we employed a double machine learning approach to evaluate the policy effects of MEC on sustainable resource development. We constructed an interactive double machine learning model to examine and validate the specific mechanisms of resource development effects. The results indicate that the level of sustainable resource development in MEC countries is relatively low, and a “resource curse” effect already exists. However, participating in MEC suppresses this “curse” effect. By promoting innovation cooperation, institutional improvement, structural optimization, trade openness, and pollution reduction, MEC effectively enhances the sustainable development levels of partner countries.

This paper focuses on the differences in the application of measurement and machine learning methods when making causal inferences. Newly highlighted double machine learning offers new research methods for policy or intervention evaluation in economic research panel data interpretability remains in doubt. We select the county panel data of Fujian Province, establish a quasi-natural experiment, and adopt a general double machine learning model and a differences-in-differences (DID) model to evaluate the policy effect of the new town policy on the optimisation of industrial structure, respectively. Both results demonstrate that the new town policy has a significant optimisation impact on industrial structure, but the dual machine learning results differ under this sample by the influence of the algorithm's advance selection. The stability test proves that the DID is more stable, and the economic significance is more explained, which is contrary to the premise of the universality of double machine learning.

Forecasting financial systemic risk has always been an important element of financial research. Despite being a key component of our country's financial market, the bond market has received relatively less systematic research attention from scholars in terms of singular market risk warnings. This paper draws on established research to construct a systemic risk stress index for the bond market. Innovatively, it utilizes the Empirical Mode Decomposition (EEMD) method to decompose the pressure index of the Chinese bond market from 2010 to 2023 into individual IMF sequences. Then, it employs the LSTM algorithm for ensemble forecasting, conducting systematic risk warning research. According to the simulation results, China's bond market will show a trend of declining pressure or low pressure in the long term, and the systemic risk will fluctuate less under effective regulation. Meanwhile, the EEMD-LSTM model has higher risk prediction accuracy compared with single LSTM model prediction.