人工智能图像变化检测

Posted by Jinming Qiao on March 26, 2019

人工智能图像变化检测


时间:2021-03-26 13:00:00

实现:人工智能图像变化检测方法汇总总结

姓名:乔金明


  • 孪生神经网络Siamese适合的特点(主流变换检测的网络架构,不适合水库场景)
    • 因为属于模板匹配类的算法,对于突然变化和超出边界框(图像)的目标跟踪会失败
    • 对于背景杂斑较多,即有太多相似性物体的时候,跟踪效果不好
    • 优点是淡化了标签,使得网络具有很好的扩展性,可以对那些没有训练过的类别进行分类,这点是优于很多算法的。而且这个算法对一些小数据量的数据集也适用,变相的增加了整个数据集的大小,使得数据量相对较小的数据集也能用深度网络训练出不错的效果
  • 图像变换检测当前的应用重点领域,医学图像(例如病灶变化检测),卫星图像(例如sar图像检测),街景检测。医学图像图像质量高,卫星图像具有其他的通道信息,但这两个领域对背景与无关的检测不关心,不适合水库场景。与水库场景相似的街景检测场景,可以采用相关的分析,但热度很低导致网络模型规模不大而且较老,只在实验集效果表现好,所以开发的流程网络将以changenet->CDnet->sscdnet->…的方向进行。

方法集合

  • DTCDSCN (CNN; Siamese)

  • CorrFusionNet(CNN; Scene-level; Siamese; Optical RS)

  • SSCDNet(CNN (ResNet18); Siamese; Transfer Learning; Semantic; Streetview)

  • Heterogeneous_CD(AE (Code-Aligned AE); Unsupervised; Transformation; Heterogeneous; Optical RS)

  • FDCNN(CNN (VGG16); Transfer Learning; Pure-Siamese; Multi-scale; Optical RS)

  • STANet(CNN (ResNet-18); Attention Mechanism; Pure-Siamese; Spatial–Temporal Dependency; Optical RS)

  • X-Net(CNN; Unsupervised; Transformation; Heterogeneous; Optical RS; SAR)

  • ACE-Net(AE (Adversarial Cyclic Encoders); Unsupervised; Transformation; Heterogeneous; Optical RS; SAR)

  • VGG_LR(CNN (VGG16); Transfer Learning; Pure-Siamese; SLIC; Low Ranks; Optical RS)

  • SCCN(AE (DAE); Unsupervised; Heterogeneous; Optical RS; SAR)

  • cGAN(GAN (conditional GAN); Heterogeneous; Optical RS; SAR)

  • DASNet(CNN (VGG16); Siamese; Attention Mechanism ; Optical RS)

  • UNetLSTM(CNN (UNet); RNN (LSTM); Integrated Model; Optical RS)

  • CDMI-Net(CNN (Unet); Pure-Siamese; Multiple Instance Learning; Landslide Mapping; Optical RS)

  • DSFANet(DNN; Unsupervised; Pre-classification; Slow Feature Analysis; Optical RS)

  • CD-UNet++(CNN (improved UNet++); Direct Classification; Optical RS)

  • SiameseNet(CNN (VGG16); Pure-Siamese; Optical RS)

  • FC-EF, FC-Siam-conc, FC-Siam-diff(CNN (UNet); Pure-Siamese; Optical RS)

  • CosimNet(CNN (Deeplab v2); Pure-Siamese; Streetview)

  • Mask R-CNN(Mask R-CNN (ResNet-101); Transfer Learning; Post-Classification; Optical RS)

  • CaffeNet(CNN (CaffeNet); Unsupervised; Transfer Learning; Optical RS)

  • CWNN(CNN (CWNN); Unsupervised; Pre-Classification; SAR)

  • MLFN(CNN (DenseNet); Transfer learning; SAR)

  • GarborPCANet(CNN (PCANet); Unsupervised; Pre-Classification; Gabor Wavelets; SAR)

  • Ms-CapsNet(CNN (Ms-CapsNet); Capsule; Attention Mechanism; Adaptive Fusion Convolution; SAR)

  • DCNet(CNN; Unsupervised; Pre-Classification; SAR)

  • ChangeNet(CNN; Siamese; StreetView)