Foreign Object Detection in Railway Images Based on an Efficient Two-Stage Convolutional Neural Network

Figure 2

Overall architecture of the proposed railway image classification network. Each rectangle block represents a layer. The operations below the blocks are operations conducted on that layer, and the figures on top of the blocks are the output feature size of that layer. The IIRU-SK is the improved inverted residual unit combined with the selective kernel convolution, and IIRU-CBAM is the improved inverted residual unit combined with the convolutional block attention module.