19 October 2020 in IEEE Access
Existing works on leather surface defects are mainly focused on the defect detection without performing the classification automatically. On the other hand, the manual classification has the disadvantages including a misjudgment, the occurrence of cumbersome labors and a high cost. To address these issues, this paper proposes a framework for classifying the leather surface defects based on a parameter optimized residual network. In this framework, an ultra high definition imaging system is first utilized to obtain the images of the leather surface. Then, two important network parameters are optimized. They are the size of the image data set and the size of the sliding patch window. Here, the size of the image data set is determined by achieving the tradeoffs between the evaluated workload and the classification accuracy. On the other hand, the size of the sliding patch window is obtained by the least squares method. In order to mitigate the limitation of the gradient disappearance in a deep network, a residual module is introduced to the proposed framework. Finally, the defect classification is performed based on the extracted features using multi-layer convolution and pooling operations. Computer numerical simulation results show that the leather surface defects can be classified effectively by the proposed framework. Also, the classification accuracy can reach 94.6%.
26 March 2025 in IEEE Transactions on Emerging Topics in Computational Intelligence
Cross-database facial expression recognition (CDFER) has attracted increasing attention when evaluating the systems' generalization performance. Although the attention mechanism can capture the feature-wise importance or feature-correlation of expression sensitive regions, the attention-based network suffers from the overfitting to the source database, due to possible over-dependence on most salient features, without exploring feature characteristics during removal of feature redundancy. To address this issue, this paper introduces a multi-kernel competitive convolution in feature-wise attention to obtain more salient regions and let each kernel compete with others to enhance the expressive ability of features, by reducing attention overfitting to the source domain. For feature-correlation attention, we resort to a Monte Carlo-based dropout to not only reduce the over-learning of the feature relationship, but also model the dropout probability distribution more specifically by taking the characteristics of feature maps into account. Experimental results show that our algorithm achieves much better generalization performances than the state of the arts (SOTAs) on six publicly available datasets, in the scenarios of single source domain, multiple source domains and domain adaption.