Aiming at the problem of a large number of small dense objects in high-altitude shooting and complex background noise interference in the captured scenes, an improved object detection algorithm for YOLOv5 UAV capture scenes is proposed. A Feature Enhancement Block (FEBlock) is first proposed to generate adaptive weights for different receptive field features by convolution, assigning major weights to shallow feature maps to improve small object feature extraction ability. The FEBlock is then integrated into Spatial Pyramid Pooling (SPP) to generate Enhanced Spatial Pyramid Pooling (ESPP), which performs feature enhancement for the result of each maximum pooling; and creates new features containing multi-scale contextual information with better feature characterization capability by weighting fused contextual features. Secondly, the Self-Characteristic Expansion Plate (SCEP) is proposed, which achieves the fusion and expansion of feature information through compression, non-linear mapping, and expansion with its own module, further improving the network’s capacity for feature extraction and generating a new spatial pyramid pooling (ESPP-S) by splicing with ESPP. Finally, a shallower feature map is added as a detection layer to the YOLOv5 network model’s large, medium, and small detection layers to improve the network’s detection performance for medium and long-range objects. Experiments were conducted on the VisDrone2021 dataset, and the results showed that the improved YOLOv5 model improved mAP0.5 by 4.6%, mAP0.5:0.95 by 2.9%, and precision by 2.7%. The mAP0.5 of the model trained at the input resolution of 1024 × 1024 reached 56.8%. The experiments show that the improved YOLOv5 model can improve object detection accuracy for UAV capture scenes.
1 August 2015 in IET Control Theory & Applications
Identification and composite control of Hammerstein system with unknown order linear dynamics and a static hysteresis non‐linearity modelled by Preisach operator is investigated. The order of linear dynamics is firstly determined by Hankel matrix approach and then a blind identification is implemented to identify the linear dynamics based on the over‐sampling output measurements only. Then a novel deterministic approach is proposed to identify the Preisach model for hysteresis non‐linearity, which is devoted to identify a triangle matrix. This novel approach needs less dimensions to obtain Preisach density function than other existing methods. Finally, a composite control consisting of discrete inverse model‐based controller (DIMBC) and discrete adaptive sliding mode controller (DASMC) is developed to achieve tracking control. The composite control can reduce the reaching time of DASMC and improve the robustness of DIMBC. Experiments based on a turntable servo system demonstrate the effectiveness of the proposed identification and control methods.
This paper is concerned with the problem of personnel localization in the complex coal mine environment with wireless channel fading and unknown noise statistics. Considering the random channel fading caused by signal fluctuation and transmission fault, an improved adaptive unscented Kalman filter (IAUKF) algorithm is proposed. The mean and error covariances of noise are estimated adaptively by adopting the improved Sage–Husa noise estimation method. In order to save energy and improve energy utilization, the multi-sensor clustering is performed to divide the spatial distribution of sensors into multiple clusters. The sensors in the same cluster can communicate with each other to maintain the consistency of estimation. The simulation results show that the IAUKF algorithm is better than extended Kalman filter (EKF), unscented Kalman filter (UKF), and improved unscented Kalman filter (IUKF) algorithms.
An adaptive high‐order neural network (HONN) control strategy is proposed for a hysteresis motor driving servo system with the Bouc‐Wen model. To simplify control design, the model is rewritten as a canonical state space form firstly through coordinate transformation. Then, a high‐gain state observer (HGSO) is proposed to estimate the unknown transformed state. Afterward, a filter for the tracking errors is adopted which converts the vector error e into a scalar error s. Finally, an adaptive HONN controller is presented, and a Lyapunov function candidate guarantees that all the closed‐loop signals are uniformly ultimately bounded (UUB). Simulations verified the effectiveness of the proposed neural network adaptive control strategy for the hysteresis servo motor system.
4 January 2020 in Mathematical Problems in Engineering
An echo state network (ESN) for extended state observer (ESO) and sliding mode control (SMC) of permanent magnet synchronous motor (PMSM) in an electric vehicle system is investigated in this paper. For the PMSM model, most researches neglect the hysteresis loss and other nonlinear factors, which reduces the accuracy of the PMSM model. We present a modified PMSM model considering the hysteresis loss and then transform the new PMSM model to a canonical form to simplify the controller design. In order to deal with the hysteresis loss, an ESN is utilized to estimate the nonlinearity. Considering that some states cannot be directly obtained, an ESO with ESN is proposed to estimate unknown system states of the electric vehicle PMSM system. Afterwards, an SMC is adopted to control the closed-loop system based on the ESO with ESN, and a double hyperbolic function instead of the sign function is used to suppress the chattering of the SMC. The stabilities of the observer and the controller are all guaranteed by Lyapunov functions. Finally, simulations are presented to verify the validity of the echo state network for extended state observer and the neural network sliding mode control.