An efficient fusion algorithm for the rotary-wing flying robot is presented for solving path planning problem in the 3D mountain environment. This fusion algorithm combines A*algorithm with artificial potential field method both improved and optimized for 3D environment. Firstly, A*algorithm is adopted to plan the preliminary path in the environmental model established by grid method. This A* algorithm is optimized in the heuristic function according to the 3D environment. Then, we select and save the key nodes in the preliminary path. Finally artificial potential field algorithm is used to do the path smoothing processing on the basis of these key nodes. We carried out the simulation in MATLAB of the proposed algorithm, single A*algorithm and artificial potential field method respectively in the 3D map consisting of random irregular surface and defined peak. The simulation results indicates that the fusion algorithm is feasible in 3D path planning and superior in smoothness performance, which can satisfy the property of real-time tracking for rotary-wing flying robots.
27 September 2017 in International Journal of Parallel Programming
This article proposes a clothing image retrieval method based on the ConvNeXt network. This research uses the ConvNeXt network, which is obtained by improving and adjusting the classic ResNet50 network based on some advanced ideas of the transformer network, as a feature extractor to conduct research on a subset (In-Shop Clothes Retrieval) of the DeepFashion dataset. In order to improve model training speed and retrieval performance, this article uses transfer learning, which utilizes the ConvNeXt network model pre-trained on the ImageNet_1k dataset. Considering the limitations of using only a global branch, a local branch is added, and the global branch and local branch are trained together during training. During retrieval, only global features are used, as local features help global features learn some knowledge of the local features, resulting in better results. Experimental results show that this method can extract more expressive clothing image features and effectively improve clothing image retrieval performance.