For the challenge of few-shot sales prediction in supply chain applications, traditional methods often struggle with data scarcity and fail to effectively leverage related information across multiple domains. In response, this study introduces a novel Graph Attention Enhanced Multi-Scale Transfer Learning (GAT-MSTL) method to enhance predictive accuracy in small-sample sales forecasting. GAT-MSTL constructs temporal similarity graphs using Graph Attention Networks (GAT) to dynamically capture cross-domain temporal dependencies and performs multi-scale source domain pre-training to facilitate effective knowledge transfer. Our experimental validation on the Corporación Favorita Grocery Sales Forecasting dataset demonstrates that GAT-MSTL outperforms various existing methods, delivering superior average forecasting, highlighting its effectiveness in addressing data scarcity problem in supply chain sales forecasting.

The first scatterometer onboard Chinese meteorological satellites is a dual-frequency and dual-polarization scatterometer named Wind Radar (WindRAD). It uses an advanced fan-beam conical scanning mechanism to acquire wind vector observation of global ocean surfaces and other geophysical parameters. Since the launch in 2021, external calibration using ground-based active radar calibrator (ARC) has been carried out to evaluated WindRAD in-orbit situation and observation accuracy. Satellite-ground integrated external calibration process is proposed, and optimal satellite-ground observation mode is specially designed for the WindRAD, greatly reducing the complexity of satellite-ground interaction and improving the efficiency of external calibration observation. This article proposes the algorithm of WindRAD external calibration, with comprehensive consideration of scientific nature and engineering realizability. WindRAD in-orbit observation data as well as ARC data were used to calculate the real antenna patterns and absolute calibration coefficients of each polarization for both C- and Ku-bands. The evaluation results revealed that the in-orbit antenna pattern hadn’t changed much compared with the prelaunch test result, which for the first time confirmed the correctness of the key parameters used in WindRAD calibration. Moreover, the calibration accuracy is better than 1 dB. For the first time, the stability of WindRAD in orbit is confirmed using external active reference target.

Deep Convolutional Neural Networks (CNNs) have demonstrated excellent performance in various multimedia application scenarios. However, complex models often require significant computational resources and energy costs. Therefore, CNN compression is crucial for addressing deployment challenges of multimedia application on resource constrained edge devices. However, existing CNN channel pruning strategies primarily focus on the "weights" or "activations" of the model, overlooking its "interpretability" information. In this paper, we explore CNN pruning strategies from the perspective of model interpretability. We model the correspondence between channel feature maps and interpretable visual perception based on class saliency maps, aiming to assess the contribution of each channel to the desired output. Additionally, we utilize Discrete Wavelet Transform (DWT) to capture the global features and structure of class saliency maps. Based on this, we propose a Channel Spatial Dependability (CSD) metric, evaluating the importance and contribution of channels in a bidirectional manner to guide model pruning. And we dynamically adjust the pruning rate of each layer based on performance changes, in order to achieve more accurate and efficient adaptive pruning. Our method achieves significant results across a range of different networks and datasets. For instance, we achieved a 51.3% pruning on the ResNet-56 model while maintaining an accuracy of 94.16%, outperforming feature-map or other State-of-the-Art (SOTA).

The first C- and Ku-band dual-frequency scatterometer (WindRAD) onboard the Chinese FengYun-3E (FY-3E) satellite was successfully launched in July 2021. The WindRAD scatterometer uses an advanced fan-beam conical scanning mechanism to acquire wind vector data of global ocean surfaces and other geophysical parameters through backscattering measurements of the Earth. This article provides an introduction to the WindRAD instrument, an overview of the data preprocessing, and assessment of WindRAD measurements. The numerical weather prediction-based ocean calibration (NOC) approach, natural targets, and cross-calibration against the Ku-band scatterometer onboard the HY-2B satellite based on collocated backscatter measurements, were used to validate WindRAD backscatter results. The evaluation results revealed that the performance of the WindRAD data is generally in good agreement with other scatterometer data currently in use, while WindRAD backscatter data may contain nonlinear calibration issues that require further investigation. WindRAD has the ability to provide high-quality global backscattering measurements, which can be used for the inversion of various geophysical parameters and assimilation applications.