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
注
- 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)