Search

The Elman neural network has good dynamic properties and strong global stability, being most widely used to deal with nonlinear, dynamic, and complex data. However, as an optimization of the backpropagation (BP) neural network, the Elman model inevitably inherits some of its inherent deficiencies, influencing the recognition precision and operating efficiency. Many improvements have been proposed to resolve these problems, but it has proved difficult to balance the many relevant features such as storage space, algorithm efficiency, recognition precision, etc. Also, it is difficult to obtain a permanent solution from a temporary solution simultaneously. To address this, a genetic algorithm (GA) can be introduced into the Elman algorithm to optimize the connection weights and thresholds, which can prevent the neural network from becoming trapped in local minima and improve the training speed and success rate. The structure of the hidden layer can also be optimized using the GA, which can solve the difficult problem of determining the number of neurons. Most previous studies on such evolutionary Elman algorithms optimized the connection weights or network structure individually, which represents a slight deficiency. We propose herein a novel optimized GA-Elman neural network algorithm where the connection weights are real-encoded, while the neurons of the hidden layer also adopt real-coding but with the addition of binary control genes. In this new algorithm, the connection weights and the number of hidden neurons are optimized using hybrid encoding and evolution simultaneously, greatly improving the performance of the resulting novel GA-Elman algorithm. The results of three experiments show that this new GA-Elman model is superior to the traditional model in terms of all calculated indexes. [ABSTRACT FROM AUTHOR]

• The attention modules are constructed to extract and fuse RGB-D data effectively. • A fusion branch and the bi-direction multi-step propagation strategy are introduced. • Extensive experiments are performed on public NYUDv2 and SUN-RGBD datasets. Semantic segmentation based on the complementary information from RGB and depth images has recently gained great popularity, but due to the difference between RGB and depth maps, how to effectively use RGB-D information is still a problem. In this paper, we propose a novel RGB-D semantic segmentation network named RAFNet, which can selectively gather features from the RGB and depth information. Specifically, we construct an architecture with three parallel branches and propose several complementary attention modules. This structure enables a fusion branch and we add the Bi-directional Multi-step Propagation (BMP) strategy to it, which can not only retain the feature streams of the original RGB and depth branches but also fully utilize the feature flow of the fusion branch. There are three kinds of complementary attention modules that we have constructed. The RGB-D fusion module can effectively extract important features from the RGB and depth branch streams. The refinement module can reduce the loss of semantic information and the context aggregation module can help propagate and integrate information better. We train and evaluate our model on NYUDv2 and SUN-RGBD datasets, and prove that our model achieves state-of-the-art performances. [ABSTRACT FROM AUTHOR]