基于轻量级卷积神经网络的羊绒羊毛识别方法

Cashmere and wool identification method based on lightweight convolutional neural network

羊绒、羊毛纤维的形态和物理化学性质十分相似,2种纤维表面鳞片的纹理有所不同,鉴别二者的传统方法显微镜人工鉴别存在速度慢、识别率不高、人力成本高等弊端。针对该问题,文章提出了一种基于轻量级卷积神经网络MobileNetV3_small模型的纤维识别方法。实验发现:纤维图像中的鳞片纹理模式复杂度有限,轻量级网络能够有效地提取纤维图像中的视觉特征,并根据特征较好地识别出纤维的类别,实验中5种不同的纤维测试集识别率超过97.1%。与其他卷积神经网络相比,轻量级模型MobileNetV3_small速度更快,识别5000个样本只需13 s,适合于纤维商检中的快速检测。

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 5000 samples,which makes it highly suitable for rapid fiber detection applications,especially in commercial inspections.