基于改进MobileNetV2的棉花颜色分级检测  

Cotton color grade detection based on improved MobileNetV2

作　　者：王中璞 吴正香 尤美路 张立杰[1] 阿不都热西提·买买提 WANG Zhongpu;WU Zhengxiang;YOU Meilu;ZHANG Lijie;ABUDUREXITI Maimaiti(Xinjiang University,Urumqi,830046,China;Fiber Quality Monitoring Center of Xinjiang Uygur Autonomous Region,Urumqi,830046,China)

机构地区：[1]新疆大学,新疆乌鲁木齐830046 [2]新疆维吾尔自治区纤维质量监测中心,新疆乌鲁木齐830026

出　　处：《棉纺织技术》2024年第6期15-21,共7页Cotton Textile Technology

基　　金：新疆维吾尔自治区科技重大专项(2022A01008-1)。

摘　　要：针对棉花颜色级检验中感官检验容易受到主观因素影响、仪器检验不稳定的问题,提出一种使用改进MobileNetV2神经网络实现棉花颜色级检测的方法。通过自主设计的图像采集装置,收集白棉一级到白棉五级5种棉花颜色级样品,制作数据集。将MobileNetV2网络后三层进行特征融合,并嵌入CBAM注意力机制,同时与GhostNet、ShuffleNetV2和原始MobileNetV2模型进行对比,预测棉花颜色分级。结果表明:改进后的MobileNetV2在测试集的准确率达到92.10%,相对于GhostNet、ShuffleNetV2和原始MobileNetV2分别提高了3.01个百分点、4.61个百分点、1.24个百分点,具有较好的检测效果。In order to solve the problems that sensory inspection was easily influenced by subjective factors and unstable instrument inspection in cotton color grade detection,a method of using improved MobileNetV2 neural network to realize cotton color grade detection was proposed.Through the self-designed image acquisition device,5 kinds of cotton color samples from white cotton grade 1 to white cotton grade 5 were collected,and the data set was made.The features of the last three layers of MobileNetV2 network were fused and embedded into CBAM attention mechanism.Meanwhile,it was compared with GhostNet,ShuffleNetV2 and the original MobileNetV2 model to predict cotton color grade.The results showed that the improved MobileNetV2 accuracy in the test set was achieved 92.10%,which was 3.01 percentage points higher than GhostNet,4.61 percentage points higher than ShuffleNetV2,and 1.24 percentage points higher than the original mobileNetV2.It has better detection effect.

关 键 词：MobileNetV2模型 棉花颜色级 神经网络 注意力机制 特征融合 

分 类 号：TS111.9[轻工技术与工程—纺织材料与纺织品设计]