基于改进的AlexNet网络的服装廓形识别

Garment Silhouette Recognition Based on Improved AlexNet Network

为了提高服装廓形的识别准确性,实现平面款式图的自动分类和识别,提出了一种基于改进AlexNet网络的服装廓形识别算法。以女裤廓形识别为例,首先,构建了一个包含19000多张女裤平面款式图的数据集,数据集按“S”形、“A”形、“X”形、“O”形、“H”形、“V”形对样本进行标签分类,并划分为训练集、测试集和验证集;然后,构建网络模型对训练集和测试集进行训练;针对平面款式图的图像特点对AlexNet网络进行了改进,通过减小网络深度,在第4层卷积层后引入批归一化操作来防止过拟合,提高模型的泛化性;最后,采用验证集进行模型验证,运用混淆矩阵对模型的验证结果进行可视化。结果表明:改进模型在验证集上的平均准确率为88%,最高类别识别准确率为94%,比改进前的AlexNet网络的识别准确率提高2%,且相较于resnet18等其他网络而言改进后的网络准确率更高,可用于女裤廓形识别。

In order to improve the recognition accuracy of garment silhouette and realize the automatic classification and recognition of flat style drawings,this paper proposed a garment silhouette recognition algorithm based on improved AlexNet network.Taking women’s trouser silhouette recognition as an example,this study firstly constructed a dataset containing more than 19000 graphic style diagram of women s trousers.The data set was classified into training set,test set and validation set by“S”,“A”,“X”,“O”,“H”and“V”shape labels.Then the network model was constructed to train the training and test sets.The AlexNet network was improved by reducing the depth of the network and introducing a batch normalization operation after the fourth convolutional layer to prevent overfitting and improve the generalizability of the model.Finally,the validation set was used for model validation and the confusion matrix was applied to visualize the validation results of the model.The results show that the test accuracy of the improved model on the validation set is 88%,and the highest category recognition accuracy is 94%,which is 2%higher than the recognition accuracy of the AlexNet network before improvement,and the improved network has higher accuracy compared with other networks like resnet18,and can be used for women s pants silhouette recognition.