This paper presents such a comprehensive multi-scale strategy to develop realistic stochastic replicas of a composite material, with emphasis on the identification step.
The proposed multi-scale strategy is carried out by a user-material subroutine (UMAT) in ABAQUS, which predicts the detailed local responses and complex failure mechanisms of the woven composites.
Furthermore, the feasibility of complex Laplace mixed-domain inversion and multi-scale strategy is demonstrated by the 1D synthetic and 2D overthrust model tests.
This pyramidal version includes a multi-scale strategy which allows us to handle larger displacements while keeping the reduced computational load of Lucas-Kanade sparse method [22].
We evaluate the effectiveness of multi-pooling by three comparative experiments: using classical multi-scale strategy by image pyramid, using multi-pooling on single scale image and both of them, i.e., the multi-pooling of image pyramids.
The proposed multi-pooling of convolutional activation tensor can either be used alone which is very efficient or with traditional image pyramid multi-scale strategy which has better performance in terms of classification accuracy.
Receptive fields with different sizes covers different content in original image, hence, this operation is in accordance with traditional multi-scale strategy by using image pyramids, both of which will generate more comprehensive information compared with single-scale image or single activation feature map.
In this article, we propose a specific approach of such multi-scale strategy, the outline of which is to transform the constrained dictionary learning problem into the augmented Lagrangian (AL) framework first and then refine the dictionary from low scale to high.
In a designed scaling strategy, the ambition to scale is designed into the scaling strategy at its inception.
A combination of several strategies that are suited concurrently to multi-physics, multi-scale and large-size system computation is therefore required.
A multi-scale supervised neural architecture, called Multi-Scale SOON, is proposed for natural texture classification.
