Skip to content

Road extraction of high-resolution remote sensing images based on various semantic segmentation networks

Notifications You must be signed in to change notification settings

zetrun-liu/FCNs-for-road-extraction-keras

Repository files navigation

FCNs-for-road-extraction-keras

Road extraction of high-resolution remote sensing images based on various semantic segmentation networks.

Python novice, the code is redundant. Training different models requires change few lines code in some modules(main.py, sub_predict.py). Besides, data preprocessing contains images and masks, so crop_image.py also need change. The code corresponding my accepted paper, which running successful and got good segmentation results. If you have any problems while running the code, leave a comment.

Environment

Win10 + Anaconda3 + tesndorflow-gpu + keras

Main packages Required: opencv-python, scikit-image,

Details about the project

Due to FCNs can take arbitrary size image as input, however it will need amount of GPU memory to store the feature maps. Here, we utilize fixed-sized training images (256×256) to train the model. These training images are sampled from the original images by sliding windows technique.

data.py: Used as a data generator;

crop_image.py: Got samples from the original images by sliding windows technique;

model.py: Contain various FCNs model, including FCN-8s/2s, SegNet, Unet, VGGUnet, ResUnet and D-ResUnet;

metrics.py: Calculating the metrics(precision/recall/active IoU) of the predicted road segmentation maps;

sub_predict.py: In the original test images, sliding window technology with 16-pixels overlapping was used to predict each patch and splice them one by one to produce the final original size image segmentation image.

Usage

Here are the main steps of running the project:

Step1: Starting main.py to train the model and get the weights of model, which is a hdf5 type file;

Step2: Running sub_predict.py to predict the road of test data, of course you need to change few lines code for loading various model and its corresponding weights;

Step3: Using metrics.py to get the metrics of road segmentation performance.

Reference

  1. https://github.com/HLearning/unet_keras;

  2. https://github.com/zhixuhao/unet;

  3. https://github.com/DavideA/dilation-keras;

  4. https://github.com/mrgloom/awesome-semantic-segmentation

About

Road extraction of high-resolution remote sensing images based on various semantic segmentation networks

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages