A Residual Network of Water Scene Recognition Based on Optimized Inception Module and Convolutional Block Attention Module

摘要

Currently widely used Convolutional Neural Net- works(CNNs) are based on super large data sets, which have huge and complex network structure and can not be directly used in water scene recognition. For the characteristics of the water scene, a residual network based on optimized Inception module and Convolutional Block Attention Module(CBAM) is proposed in this paper. The optimized Inception module with parallel structure can extract multi-scale features, the residual structure is suitable for deeper networks, and CBAM enables the network to notice the informative parts of the images. Compared with some classical CNNs, the method has less parameters. The experimental results show that the method can effectively improve the accuracy of water scene recognition.