怎么修改网站标题关键词描述,wordpress导航菜单404,380元网站建设,东兴移动网站建设遥感地物分类通过对遥感图像中的地物进行准确识别和分类#xff0c;为资源管理、环境保护、城市规划、灾害监测等领域提供重要信息#xff0c;有助于实现精细化管理和科学决策#xff0c;提升社会治理和经济发展水平。深度学习遥感地物分类在提高分类精度、自动化程度、处理… 遥感地物分类通过对遥感图像中的地物进行准确识别和分类为资源管理、环境保护、城市规划、灾害监测等领域提供重要信息有助于实现精细化管理和科学决策提升社会治理和经济发展水平。深度学习遥感地物分类在提高分类精度、自动化程度、处理大规模数据、普适性以及推动遥感技术创新和发展等方面都具有重要的意义。本文将利用深度学习BiseNet实现遥感地物分类。 数据集 本文使用的数据集为WHDLD数据集[1](Wuhan dense labeling dataset)。WHDLD数据集包括4940张高分辨率遥感影像包含6种土地覆盖类型影像尺寸均被裁剪至256×256像素。下面是一些数据集示例。 BiSeNet BiseNet[2]Bilateral Segmentation Network是一种用于图像分割的深度学习网络。它具有双边分割的特点可以同时处理空间信息和上下文信息从而实现高效、准确的图像分割。 具体来说BiseNet由两个分支组成空间路径spatial path和上下文路径context path。其中空间路径具有较小的感受野可以捕获丰富的空间信息并生成高分辨率的特征图而上下文路径则具有较大的感受野可以捕获更多的上下文信息并生成低分辨率的特征图。这两个路径通过一个特征融合模块进行融合从而生成既包含丰富空间信息又包含上下文信息的分割结果。 在BiseNet中还有一些关键的技术和设计如轻量级模型设计、注意力机制、特征融合等这些技术和设计可以进一步提升网络的性能和效率。 网络复现 resnet18 import torchimport torch.nn as nnimport torch.nn.functional as Fimport torch.utils.model_zoo as modelzooresnet18_url https://download.pytorch.org/models/resnet18-5c106cde.pthfrom torch.nn import BatchNorm2ddef conv3x3(in_planes, out_planes, stride1): 3x3 convolution with padding return nn.Conv2d(in_planes, out_planes, kernel_size3, stridestride, padding1, biasFalse)class BasicBlock(nn.Module): def __init__(self, in_chan, out_chan, stride1): super(BasicBlock, self).__init__() self.conv1 conv3x3(in_chan, out_chan, stride) self.bn1 BatchNorm2d(out_chan) self.conv2 conv3x3(out_chan, out_chan) self.bn2 BatchNorm2d(out_chan) self.relu nn.ReLU(inplaceTrue) self.downsample None if in_chan ! out_chan or stride ! 1: self.downsample nn.Sequential( nn.Conv2d(in_chan, out_chan, kernel_size1, stridestride, biasFalse), BatchNorm2d(out_chan), ) def forward(self, x): residual self.conv1(x) residual self.bn1(residual) residual self.relu(residual) residual self.conv2(residual) residual self.bn2(residual) shortcut x if self.downsample is not None: shortcut self.downsample(x) out shortcut residual out self.relu(out) return outdef create_layer_basic(in_chan, out_chan, bnum, stride1): layers [BasicBlock(in_chan, out_chan, stridestride)] for i in range(bnum-1): layers.append(BasicBlock(out_chan, out_chan, stride1)) return nn.Sequential(*layers)class Resnet18(nn.Module): def __init__(self): super(Resnet18, self).__init__() self.conv1 nn.Conv2d(3, 64, kernel_size7, stride2, padding3, biasFalse) self.bn1 BatchNorm2d(64) self.relu nn.ReLU(inplaceTrue) self.maxpool nn.MaxPool2d(kernel_size3, stride2, padding1) self.layer1 create_layer_basic(64, 64, bnum2, stride1) self.layer2 create_layer_basic(64, 128, bnum2, stride2) self.layer3 create_layer_basic(128, 256, bnum2, stride2) self.layer4 create_layer_basic(256, 512, bnum2, stride2) self.init_weight() def forward(self, x): x self.conv1(x) x self.bn1(x) x self.relu(x) x self.maxpool(x) x self.layer1(x) feat8 self.layer2(x) # 1/8 feat16 self.layer3(feat8) # 1/16 feat32 self.layer4(feat16) # 1/32 return feat8, feat16, feat32 def init_weight(self): state_dict modelzoo.load_url(resnet18_url) self_state_dict self.state_dict() for k, v in state_dict.items(): if fc in k: continue self_state_dict.update({k: v}) self.load_state_dict(self_state_dict) def get_params(self): wd_params, nowd_params [], [] for name, module in self.named_modules(): if isinstance(module, (nn.Linear, nn.Conv2d)): wd_params.append(module.weight) if not module.bias is None: nowd_params.append(module.bias) elif isinstance(module, nn.modules.batchnorm._BatchNorm): nowd_params list(module.parameters()) return wd_params, nowd_params BiSeNet import torchimport torch.nn as nnimport torch.nn.functional as Fimport torchvisionfrom .resnet import Resnet18from torch.nn import BatchNorm2dclass ConvBNReLU(nn.Module): def __init__(self, in_chan, out_chan, ks3, stride1, padding1, *args, **kwargs): super(ConvBNReLU, self).__init__() self.conv nn.Conv2d(in_chan, out_chan, kernel_size ks, stride stride, padding padding, bias False) self.bn BatchNorm2d(out_chan) self.relu nn.ReLU(inplaceTrue) self.init_weight() def forward(self, x): x self.conv(x) x self.bn(x) x self.relu(x) return x def init_weight(self): for ly in self.children(): if isinstance(ly, nn.Conv2d): nn.init.kaiming_normal_(ly.weight, a1) if not ly.bias is None: nn.init.constant_(ly.bias, 0)class UpSample(nn.Module): def __init__(self, n_chan, factor2): super(UpSample, self).__init__() out_chan n_chan * factor * factor self.proj nn.Conv2d(n_chan, out_chan, 1, 1, 0) self.up nn.PixelShuffle(factor) self.init_weight() def forward(self, x): feat self.proj(x) feat self.up(feat) return feat def init_weight(self): nn.init.xavier_normal_(self.proj.weight, gain1.)class BiSeNetOutput(nn.Module): def __init__(self, in_chan, mid_chan, n_classes, up_factor32, *args, **kwargs): super(BiSeNetOutput, self).__init__() self.up_factor up_factor out_chan n_classes self.conv ConvBNReLU(in_chan, mid_chan, ks3, stride1, padding1) self.conv_out nn.Conv2d(mid_chan, out_chan, kernel_size1, biasTrue) self.up nn.Upsample(scale_factorup_factor, modebilinear, align_cornersFalse) self.init_weight() def forward(self, x): x self.conv(x) x self.conv_out(x) x self.up(x) return x def init_weight(self): for ly in self.children(): if isinstance(ly, nn.Conv2d): nn.init.kaiming_normal_(ly.weight, a1) if not ly.bias is None: nn.init.constant_(ly.bias, 0) def get_params(self): wd_params, nowd_params [], [] for name, module in self.named_modules(): if isinstance(module, (nn.Linear, nn.Conv2d)): wd_params.append(module.weight) if not module.bias is None: nowd_params.append(module.bias) elif isinstance(module, nn.modules.batchnorm._BatchNorm): nowd_params list(module.parameters()) return wd_params, nowd_paramsclass AttentionRefinementModule(nn.Module): def __init__(self, in_chan, out_chan, *args, **kwargs): super(AttentionRefinementModule, self).__init__() self.conv ConvBNReLU(in_chan, out_chan, ks3, stride1, padding1) self.conv_atten nn.Conv2d(out_chan, out_chan, kernel_size 1, biasFalse) self.bn_atten BatchNorm2d(out_chan) # self.sigmoid_atten nn.Sigmoid() self.init_weight() def forward(self, x): feat self.conv(x) atten torch.mean(feat, dim(2, 3), keepdimTrue) atten self.conv_atten(atten) atten self.bn_atten(atten) # atten self.sigmoid_atten(atten) atten atten.sigmoid() out torch.mul(feat, atten) return out def init_weight(self): for ly in self.children(): if isinstance(ly, nn.Conv2d): nn.init.kaiming_normal_(ly.weight, a1) if not ly.bias is None: nn.init.constant_(ly.bias, 0)class ContextPath(nn.Module): def __init__(self, *args, **kwargs): super(ContextPath, self).__init__() self.resnet Resnet18() self.arm16 AttentionRefinementModule(256, 128) self.arm32 AttentionRefinementModule(512, 128) self.conv_head32 ConvBNReLU(128, 128, ks3, stride1, padding1) self.conv_head16 ConvBNReLU(128, 128, ks3, stride1, padding1) self.conv_avg ConvBNReLU(512, 128, ks1, stride1, padding0) self.up32 nn.Upsample(scale_factor2.) self.up16 nn.Upsample(scale_factor2.) self.init_weight() def forward(self, x): feat8, feat16, feat32 self.resnet(x) avg torch.mean(feat32, dim(2, 3), keepdimTrue) avg self.conv_avg(avg) feat32_arm self.arm32(feat32) feat32_sum feat32_arm avg feat32_up self.up32(feat32_sum) feat32_up self.conv_head32(feat32_up) feat16_arm self.arm16(feat16) feat16_sum feat16_arm feat32_up feat16_up self.up16(feat16_sum) feat16_up self.conv_head16(feat16_up) return feat16_up, feat32_up # x8, x16 def init_weight(self): for ly in self.children(): if isinstance(ly, nn.Conv2d): nn.init.kaiming_normal_(ly.weight, a1) if not ly.bias is None: nn.init.constant_(ly.bias, 0) def get_params(self): wd_params, nowd_params [], [] for name, module in self.named_modules(): if isinstance(module, (nn.Linear, nn.Conv2d)): wd_params.append(module.weight) if not module.bias is None: nowd_params.append(module.bias) elif isinstance(module, nn.modules.batchnorm._BatchNorm): nowd_params list(module.parameters()) return wd_params, nowd_paramsclass SpatialPath(nn.Module): def __init__(self, *args, **kwargs): super(SpatialPath, self).__init__() self.conv1 ConvBNReLU(3, 64, ks7, stride2, padding3) self.conv2 ConvBNReLU(64, 64, ks3, stride2, padding1) self.conv3 ConvBNReLU(64, 64, ks3, stride2, padding1) self.conv_out ConvBNReLU(64, 128, ks1, stride1, padding0) self.init_weight() def forward(self, x): feat self.conv1(x) feat self.conv2(feat) feat self.conv3(feat) feat self.conv_out(feat) return feat def init_weight(self): for ly in self.children(): if isinstance(ly, nn.Conv2d): nn.init.kaiming_normal_(ly.weight, a1) if not ly.bias is None: nn.init.constant_(ly.bias, 0) def get_params(self): wd_params, nowd_params [], [] for name, module in self.named_modules(): if isinstance(module, nn.Linear) or isinstance(module, nn.Conv2d): wd_params.append(module.weight) if not module.bias is None: nowd_params.append(module.bias) elif isinstance(module, nn.modules.batchnorm._BatchNorm): nowd_params list(module.parameters()) return wd_params, nowd_paramsclass FeatureFusionModule(nn.Module): def __init__(self, in_chan, out_chan, *args, **kwargs): super(FeatureFusionModule, self).__init__() self.convblk ConvBNReLU(in_chan, out_chan, ks1, stride1, padding0) ## use conv-bn instead of 2 layer mlp, so that tensorrt 7.2.3.4 can work for fp16 self.conv nn.Conv2d(out_chan, out_chan, kernel_size 1, stride 1, padding 0, bias False) self.bn nn.BatchNorm2d(out_chan) # self.conv1 nn.Conv2d(out_chan, # out_chan//4, # kernel_size 1, # stride 1, # padding 0, # bias False) # self.conv2 nn.Conv2d(out_chan//4, # out_chan, # kernel_size 1, # stride 1, # padding 0, # bias False) # self.relu nn.ReLU(inplaceTrue) self.init_weight() def forward(self, fsp, fcp): fcat torch.cat([fsp, fcp], dim1) feat self.convblk(fcat) atten torch.mean(feat, dim(2, 3), keepdimTrue) atten self.conv(atten) atten self.bn(atten) # atten self.conv1(atten) # atten self.relu(atten) # atten self.conv2(atten) atten atten.sigmoid() feat_atten torch.mul(feat, atten) feat_out feat_atten feat return feat_out def init_weight(self): for ly in self.children(): if isinstance(ly, nn.Conv2d): nn.init.kaiming_normal_(ly.weight, a1) if not ly.bias is None: nn.init.constant_(ly.bias, 0) def get_params(self): wd_params, nowd_params [], [] for name, module in self.named_modules(): if isinstance(module, (nn.Linear, nn.Conv2d)): wd_params.append(module.weight) if not module.bias is None: nowd_params.append(module.bias) elif isinstance(module, nn.modules.batchnorm._BatchNorm): nowd_params list(module.parameters()) return wd_params, nowd_paramsclass BiSeNetV1(nn.Module): def __init__(self, n_classes, aux_modetrain, *args, **kwargs): super(BiSeNetV1, self).__init__() self.cp ContextPath() self.sp SpatialPath() self.ffm FeatureFusionModule(256, 256) self.conv_out BiSeNetOutput(256, 256, n_classes, up_factor8) self.aux_mode aux_mode if self.aux_mode train: self.conv_out16 BiSeNetOutput(128, 64, n_classes, up_factor8) self.conv_out32 BiSeNetOutput(128, 64, n_classes, up_factor16) self.init_weight() def forward(self, x): H, W x.size()[2:] feat_cp8, feat_cp16 self.cp(x) feat_sp self.sp(x) feat_fuse self.ffm(feat_sp, feat_cp8) feat_out self.conv_out(feat_fuse) if self.aux_mode train: feat_out16 self.conv_out16(feat_cp8) feat_out32 self.conv_out32(feat_cp16) return feat_out, feat_out16, feat_out32 elif self.aux_mode eval: return feat_out, elif self.aux_mode pred: feat_out feat_out.argmax(dim1) return feat_out else: raise NotImplementedError def init_weight(self): for ly in self.children(): if isinstance(ly, nn.Conv2d): nn.init.kaiming_normal_(ly.weight, a1) if not ly.bias is None: nn.init.constant_(ly.bias, 0) def get_params(self): wd_params, nowd_params, lr_mul_wd_params, lr_mul_nowd_params [], [], [], [] for name, child in self.named_children(): child_wd_params, child_nowd_params child.get_params() if isinstance(child, (FeatureFusionModule, BiSeNetOutput)): lr_mul_wd_params child_wd_params lr_mul_nowd_params child_nowd_params else: wd_params child_wd_params nowd_params child_nowd_params return wd_params, nowd_params, lr_mul_wd_params, lr_mul_nowd_params 训练过程精度变化 测试精度 结果展示 总结 今天的分享到此结束感兴趣的点点关注后续将分享更多案例。 参考资料 [1] WHDLD: https://sites.google.com/view/zhouwx/dataset#h.p_hQS2jYeaFpV0 [2] BiSeNet: https://arxiv.org/abs/1808.00897 本文由 mdnice 多平台发布
