卷积神经网络支持下高相似度图像识别技术研究

打开文本图片集

中图分类号：TP391.41 文献标志码：A 文章编号：2095-2945（2025）27-0050-05

Abstract：ThispaperproposesanimprovedConvolutionalNeural Network（CNN）forhighsimilarityimagerecognition.By integratingmulti-scaleandasymmetricconvolutionalkermelswithResidualNetwork（ResNet）structures，featureextractionis optimized.Dataaugmentationandregularizationenhancemodelrobustnes.Experimentson CIFAR-1OandImageNetsubsetsshow thattheimprovedCNNoutperformstraditionalCNNsandhand-craftedfeaturemethods，efectivelycapturingsubtledierencesin similar images.

Keywords：highsimilarityimagerecognition;ConvolutionalNeuralNetwork；featureextraction；dataaugmentation;residual network

高相似度图像识别在刑侦、医学影像等领域有着广泛应用，但图像间细微差异使得传统手工特征方法难以满足高精度需求。（剩余6006字）