Particle image velocimetry (PIV) plays a significant role in experimental fluid mechanics, which aims to extract the velocity fields from successive particle image pairs. Deep learning (DL) techniques have been proposed to solve such a fluid motion estimation problem. However, the existing DL methods put emphasis on accuracy while ignoring the problem of model redundancy and low computational efficiency. In this article, we propose a novel lightweight convolutional neural network called LightPIVNet for PIV estimation, which is targeted to improve the advanced optical flow model Recurrent All-Pairs Field Transforms (RAFT). Furthermore, considering the real fluid scene, we generated a new dataset PIV-Dataset-II to train our network, which increases the amount and variety of flow fields. Our approach has been verified and analyzed on synthetic and experimental particle images. The experimental results indicate that our method achieves state-of-the-art performance among all the DL methods and is much superior to those classical traditional methods, such as WIDIM and HS optical flow. Meanwhile, our model LightPIVNet reduces parameters by 40.2% and improves inference time by 14.3% compared to the current best DL model PIV-LiteFlowNet-en.

In frequency division duplex model, the massive multiple-input multiple-output (MIMO) systems rely on feedback of channel state information (CSI) to perform precoding operations. This can increase the potential transmission gain of the system. The use of large scale antennas increases the feedback overhead of CSI exponentially. Therefore, CSI must be compressed for feedback. However, convolution based feedback methods lack long-range dependency modeling of the CSI, resulting in limited effects at high compression ratios. In this letter, we propose a novel neural network based on the convolutional transformer architecture to improve the performance of CSI compression and reconstruction at high compression rates. The experimental results show that the average accuracy of ours is improved by 5.39% over the state-of-the-art method at a high compression rate of 1/64. At the same time, the overall performance of the system has been improved.

In underwater robots, accurate identification of small fish is still a major challenge, because small fish move faster and occupy less screen space, which requires higher detection flexibility and receptive field of the model. To solve this challenge, we propose a high-precision small fish identification and tracking method named YOLO8-FASG in this paper. Specifically, the proposed method is improved in three aspects based on the YOLOv8 framework. First, Alterable Kernel Convolution(AKConv) is used in the neck network of the model to automatically adjust the shape of the convolution kernel according to the size and shape of the object. In this way, the shape and contour characteristics of rapidly changing fish can be captured more accurately and efficiently; Second, we introduce a global attention mechanism (GAM) to broaden the receptive field of the model by enhancing attention to fish features from the two dimensions of channel and space; Third, we employ Simplified Spatial Pyramid Pooling-Fast(SimSPPF) to replace the standard Spatial Pyramid Pooling-Fast(SPPF) to enhance prediction accuracy. These improvements enable the model to effectively extract image features of small, fast fish, thereby improving the robot’s accuracy in identifying small fish underwater. Experiments results in the public dataset Fish4Knowledge show that YOLO8-FASG performs significantly better than traditional YOLOv8 in underwater environments. Specifically, Precision and Recall increased by 1.6% and 3.5% respectively, while mAP50 and mAP50-95 increased by 1.3% and 6.1% respectively, and our method provides an effective solution for underwater robots to identify fish schools.

A wide range of research problems in physics and engineering involve the acquisition of high-resolution data. Recently, deep learning has proved to be a prospective technique for super-resolution (SR) reconstruction of fluid flows. General deep learning methods develop temporal multi-branch networks to improve SR accuracy while ignoring computational efficiency. Further, the generalization ability of the deep learning model in different fluid flow scenarios is still an unstudied issue. In this article, we propose an efficient multi-scale integration network called FlowSRNet to reconstruct the high-resolution flow fields. Specifically, we elaborately design a lightweight multi-scale aggregation block (LMAB) to capture multi-scale features of fluid data, which contains a parallel cascading architecture and feature aggregation module. The residual backbone of FlowSRNet is built by cascading the LMABs (cascaded blocks number N = 8) in a serial manner. Also, we present a small architecture LiteFlowSRNet (cascaded blocks number N = 2) for comparison. In addition, a corresponding SR dataset is constructed to train and test the proposed model, which contains different kinds of fluid flows. Finally, extensive experiments are performed on different fluid data to evaluate the performance of the proposed model. The results demonstrate that our approach achieves state-of-the-art SR performance on various fluid flow fields. Notably, our method enjoys merit of lightweight, which facilitates the development of the complicated calculation in computational fluid dynamics.