Recent advances in deep learning have enabled state-of-the-art performance in detecting medium and large-size objects. However, small object detection remains challenging primarily due to the scarcity of information. This paper proposes an end-to-end fusion network that integrates deep and hand-crafted features to address this limitation. A fusion module based on semantic context information is designed to enhance feature discrimination ability. Additionally, we introduce a kind of feature-contrast loss to incorporate prior knowledge into the learning of deep feature according to contrastive learning. Experiments on the MS COCO dataset (34.4% AP S ) demonstrate that our approach achieves improved detection accuracy over previous methods, especially for small objects.
This paper addresses image enhancement and 3D reconstruction techniques for dim scenes inside the vacuum chamber of a nuclear fusion reactor. First, an improved multi-scale Retinex low-light image enhancement algorithm with adaptive weights is designed. It can recover image detail information that is not visible in low-light environments, maintaining image clarity and contrast for easy observation. Second, according to the actual needs of target plate defect detection and 3D reconstruction inside the vacuum chamber, a defect reconstruction algorithm based on photometric stereo vision is proposed. To optimize the position of the light source, a light source illumination profile simulation system is designed in this paper to provide an optimized light array for crack detection inside vacuum chambers without the need for extensive experimental testing. Finally, a robotic platform mounted with a binocular stereo-vision camera is constructed and image enhancement and defect reconstruction experiments are performed separately. The results show that the above method can broaden the gray level of low-illumination images and improve the brightness value and contrast. The maximum depth error is less than 24.0% and the maximum width error is less than 15.3%, which achieves the goal of detecting and reconstructing the defects inside the vacuum chamber.
Overcoming the high cost of self-attention operation in Transformer-based object detection methods and improving the detection accuracy of small objects is one of the difficulties in the field of object detection research. This paper designs a progressive assisted object detection method PaoDet based on Transformer, which uses common feature extraction backbone such as Resnet and ViT to extract multi-scale features of the input image, and uses RPN(Region Proposal Network) to extract proposals of different scales; Subsequently, a progressive modeling approach was adopted to perform self-attention and cross-attention operations on proposals of different scales from large to small, achieving feature interaction between instances, ensuring high detection efficiency and low computational complexity. During the training process, each layer of the network has certain generalization ability for detecting adjacent scale objects under the supervision of a dynamic scale division method. Compared with state-of-the-art object detection methods on COCO and UAVDT datasets, the effectiveness and superiority of the proposed method were demonstrated.
The CFETR multi-purpose overload robot (CMOR) is a key subsystem of the remote handling system of the China fusion engineering test reactor (CFETR). This paper first establishes the kinematic and dynamic models of CMOR and analyzes the working process in the vacuum chamber. Based on the uncertainty of rigid-flexible coupling, a CMOR adaptive robust sliding mode controller (ARSMC) is designed based on the Hamilton-Jacobi equation to enhance the robustness of the control system. In addition, to compensate the influence of non-geometric factors on position accuracy, an error compensation method is designed. Based on the matrix differentiation method, the CMOR coupling parameter errors are decoupled, and then the gridded workspace principle is used to identify the parameter errors and improve the motion control accuracy. Finally, the CMOR rigid-flexible coupling simulation system is established by ADAMS-MATLAB/Simulink to analyze the dynamic control effect of ARSMC. The simulation results show that the CMOR end position error exceeds for single joint motion. The average value of CMOR end position error is less than after compensation, and the absolute error value is reduced by times, improves the dynamic control accuracy of CMOR.