Cashmere and wool identification method based on lightweight convolutional neural network
Morphology and physicochemical properties of cashmere and wool fibers are highly similar,with differences in the texture of their scale patterns.Traditional identification methods for these fibers primarily rely on manual microscopic examination,which suffers from drawbacks such as slow processing speed,low recognition rates,and high labor costs.To address this issue,a fiber identification method based on the lightweight Convolutional Neural Network(CNN)model MobileNetV3_small was proposed in this paper.Experiments reveal that the complexity of scale texture patterns in fiber images is limited,and lightweight networks can effectively extract visual features from fiber images.Based on these features,the proposed method achieves superior fiber classification results.In the experiments,the recognition rates for five different fiber test sets exceed 97.1%.Compared to other convolutional neural network models,the MobileNetV3_small model exhibits faster processing speed,requiring only 13 s to identify 5 000 samples,which makes it highly suitable for rapid fiber detection applications,especially in commercial inspections.
NETL
NSTL
万方数据
维普
