The degradation of image resolution reduces the detection performance in aerial imagery because it generates a large number of small objects, and accurately detecting these small objects remains a challenge. Existing methods mostly use a Super-Resolution (SR) model to first obtain the SR image of the Low-Resolution degraded image ( $I^{LR}$ ) and then use this image as the input of the object detection (OD) network to solve this problem. However, this architecture that involves executing a complex SR network before the detector is time-consuming and makes it hard to achieve real-time model inference. To address this challenge, we propose a simple and effective rotated small object detection method, named End-to-end Super-Resolution enhanced real-Time rotated object Detector (ESRTMDet). Firstly, we design a lightweight Embedded feature map Super-Resolution Module (ESRM) embedded in the detection model to enhance and amplify the backbone output features, making the detection heads detect small objects more easily. Furthermore, we train a parallel SR network branch (PSRB) simultaneously that uses the backbone feature to restore a High-Resolution image ( $I^{HR}$ ). Through our proposed feature alignment loss ( $\mathcal {L}_{AL}$ ) and feature affinity layer (FAL), our PSRB effectively guides the feature map enhancement of ESRM. Finally, through end-to-end joint optimization of the detector and PSRB, the detection performance on $I^{LR}$ is significantly improved. Extensive experiments over DOTA and UCAS-AOD demonstrate that our method can achieve state-of-the-art (SOTA) results. Additionally, we discard our PSRB and use $I^{LR}$ as the input during inference, reducing the inference time-consuming of our model. Therefore, our ESRTMDet-X not only achieves 77.11% mean of Average Precision (mAP) on the degraded DOTA dataset, but also achieves an amazing inference speed of 337 FPS, thus obtaining the best speed-accuracy trade-off. The code and models will be released at https://github.com/liufeinuaa/aisodet.git .

Cheating has become a serious threat to the online game industry. One common type of cheating is accessing game states that are not supposed to be exposed to the players. In this paper, we evaluate a few information dissemination strategies that limit the state exposure to the client, measuring the delay introduced to the players and the system resource requirements at the game server. Our measurement is based on OpenGladiator, a multi-user, real-time strategy game that is similar to Warcraft but in open-source. We found that by performing careful on-demand preloading, we can significantly reduce the unnecessary states exposed to a client without introducing any additional delay to the players. Our measurement results show that the on-demand strategy comes with a increment to the server's CPU load. However, it also significantly reduces the server's network bandwidth consumption, which is a major cost of running a game server in the current Internet.

Traditional Euclidean spatial data processing is difficult to capture the inherent relationships of unstructured data such as bearing vibration signals. Representing vibration signals in graphical form helps to preserve their topological structure and temporal information. Secondly, most existing graph convolutional network methods are based on large graph structured data, which incurs certain memory overhead when aggregating high-order neighborhood node information and ignores important information between samples in the global graph structure. To address these issues, this paper proposes a high-order multi-head graph attention network based on an adaptive small graph structure (ASG-HOMGAT) for fault diagnosis of rolling bearings. Firstly, the adaptive preprocessing layer is used to adaptively denoise and compress the one-dimensional time-domain vibration signal, generating small rule graph data with topological structure. Then, these small graph structured data samples are input into a higher-order graph neural network, which aggregates features from multiple higher-order neighborhoods to achieve richer feature representations and fully explore the intrinsic correlation between samples. Finally, these features are aggregated into a reinforced representation of graph nodes through a multi head attention mechanism, and a SoftMax classifier is used for fault classification. The experimental results show that the ASG-HOMGAT method has better performance compared to mainstream graph neural network diagnostic models. The code and model will be released at: https://github.com/ding-ss/ASG-HOMGAT.

The intelligent fault diagnosis method based on transformer and convolutional neural network (CNN) has achieved good global and local feature extraction results. However, the multi-head self-attention mechanism adopted by the transformer and the cross-channel convolution operation in CNN increases the complexity of the model, thereby increasing the demand for hardware resources, which to some extent, limits its broad applicability in industrial applications. Therefore, this paper proposes a lightweight fault diagnosis framework based on compact multi-scale feature extraction and pruned-restructured vision transformer (ViT) to address the above challenges. Firstly, a compact multi-scale feature extraction module is designed to efficiently capture complex features in rolling bearing vibration signals through parallel multi-scale convolution kernels, combined with channel reduction strategies to significantly reduce computational complexity while maintaining feature richness. Next, short-time Fourier transform and pseudo-color processing techniques are used to obtain time–frequency images. Then, a dual optimization of matrix sparsity and structural reorganization is implemented for Self-attention in ViT to ensure model performance and significantly reduce computational overhead. Finally, the time–frequency images are segmented and rearranged before being fed into the improved lightweight ViT for global feature extraction and fault recognition of rolling bearings. The experimental results show that the proposed fault diagnosis method has the advantages of lightweight (Params:4.27 K, floating point operations per seconds:0.1 M, multiplication and accumulation operations per seconds:51.07 K) and robustness compared to mainstream algorithms.

The data-driven fault diagnosis method has achieved many good results. However, classical convolutional and recurrent neural networks have problems with large parameters and poor anti-noise performance. To solve these problems, we propose a lightweight shifted windows transformer based on inverted residual structure and residual multi-layer perceptron (IRMSwin-T) for fault diagnosis of rolling bearings. First, the original data are expanded by using overlapping sampling technology. Then, the collected one-dimensional vibration signals are vector serialized by using the patch embedding strategy. Finally, the IRMSwin-T network is developed to extract features of vector sequences and classify faults. The experimental results showed that compared with mainstream lightweight models, the IRMSwin-T model in this paper has fewer parameters and higher diagnostic accuracy.

The spectrum of data-driven fault diagnosis models is greatly expanded by deep learning. However, classical convolution and multiple branching structures have their faults in computational complexity and feature extraction. To address these issues, we propose an improved re-parameterized visual geometry group (VGG) network (RepVGG) for rolling bearing fault diagnosis. In order to meet the requirements of neural networks for the amount of data, data augmentation is performed to increase the amount of original data. Then, the original one-dimensional vibration signal is processed into a single-channel time–frequency image using the short-time Fourier transform and converted into a three-channel color time–frequency image using pseudo-color processing technology. Finally, the RepVGG model with an embedded convolutional block attention mechanism structure is developed to extract defect features from three-channel time–frequency images and perform defect classification. Two datasets of vibration data from rolling bearings are used to demonstrate the strong adaptability of this method compared with other methods.