In a super-ageing society, automating washing, which is a heavy burden among housework, helps to support the lives of elderly people. Among washing automation, the accuracy of clothing identification images obtained from a robot camera greatly influences on the subsequent operation of clothes. In this research, we aim to correctly recognize each garment from multiple clothes. We perform instance segmentation of images containing multiple clothes by Mask R - CNN which is a latest Deep Learning method. We also define the extent of overlapping of clothes as complexity and evaluate the degree to which accuracy is related to complexity.