Table 1.

Comparison of different methods for natural fiber microstructure detection.

Method	Advantages	Disadvantages
CA-DeepLabv3+	Enhanced multi-scale feature extraction through cascaded ASPP module Improved fiber detail and morphological feature perception via EMA mechanism Reduced computational complexity with lightweight MobileNetV2 Excellent handling of complex fiber interweaving and overlapping scenarios	Relatively complex multi-module structure after improvements Requires substantial annotated data for network training Relatively longer model training time
Standard DeepLabv3+	Multi-scale information capture through ASPP module Expanded receptive field via atrous convolution	High computational cost and complexity Limited feature information interaction Insufficient fiber detail extraction
U-Net	Simple encoder–decoder architecture Spatial information preservation through skip connections	Lack of dedicated multi-scale extraction mechanism Poor complex fiber structure processing Limited fiber detail perception
PSPNet	Global context capture through pyramid pooling Straightforward architecture implementation	Insufficient attention to local fiber details Basic feature fusion approach Poor fine fiber structure segmentation