基于批归一化与注意力机制的图像纹理识别算法

Image Texture Recognition Algorithm Based on Batch Normalization and Attention Mechanism

针对传统图像纹理识别方法特征提取繁琐和纹理类间模糊性高、类内区分度低的问题,提出基于批归一化与注意力机制的卷积网络图像纹理识别算法。通过逐层批归一化将分散的数据统一,优化算法损失震荡和梯度消失问题;通过通道域和空间域的注意力机制对图像的关键区域和纹理的关键特征进行突出化表达。实验结果表明,所提算法不仅参数量低、计算速度快,而且在纹理数据集上的识别率达99.84%,超越基准模型和其他网络模型,证明该算法的对图像纹理具有良好的识别效果。

Aiming at the cumbersome feature extraction of traditional image texture recognition methods,poor results,high in-ter-class ambiguity of texture,and low intra-class discrimination,a convolutional network image texture recognition algorithm based on batch normalization and attention mechanism is proposed.Through layer-by-layer batch normalization,the scattered data is unified,and the loss oscillation and gradient disappearing problems of the optimization algorithm are optimized.The key areas of the image and the key features of the texture are highlighted through the attention mechanism of the channel domain and the space domain.The experimental results show that the proposed algorithm model has low parameters and fast calculation speed.The recogni-tion rate on the dataset is 99.84%,surpassing the benchmark model and other network models,it proves that the algorithm has good recognition effect on image texture.