#MMUU-NET(MMUU-Net：A Robust and Effective Network for Farmland Segmentation of Satellite Imagery):

#Authur：Xumin Gao, Long Liu, Huaze Gong,Hangyu Zhang (Beijing Mcfly Technology Co. Ltd)

#Paper:

X. Gao, L. Liu, H. Gong. MMUU-Net: A Robust and Effective Network for Farmland Segmentation of Satellite Imagery[J]. Journal of Physics: Conference Series, 2020, 1651(1):012189 (7pp).

#Demo video:

1)https://www.bilibili.com/video/BV1Ma4y1W72W

2)https://www.bilibili.com/video/BV1fi4y1M7vU/

Requirements

Cuda 10.1
Python 3.7
Pytorch 1.3.0
Opencv 4.1.1
scipy 1.2.1
imageio 2.3.0
visdom 0.1.8.9

Usage

Data

Place 'train', 'valid' and 'test' data folders in the 'dataset' folder. [Sorry, we can't publish the data set at present]

Train

Run python train.py to train the MMUU_NET.

Predict

Run python test.py to predict on the MMUU_NET.

MMUU_Net

MMUU-Net/networks/MMUU_Net.py

Download trained MMUU-NET [百度网盘]:

链接(Link)：https://pan.baidu.com/s/1G8WhL5NEbaj-gyQOojgEoA 提取码(Password)：iat0

#Abstract

The MMUU-Net, which is a robust and effective network for satellite imagery segmentation. The encoder in the U-Net was replaced by ResNeSt, which can greatly improve classification accuracy. An ASPP layer was added in the middle. A multi-scale feature fusion module was designed in the decoder and a corresponding robust loss function was designed to improve multi- scale information fusion. Finally, in order to eliminate the adhesion phenomenon of preliminary segmentationa，a two-stage segmentation strategy including the coarse segmentation and the refined segmentation was proposed. The MIoU of MMUU-Net was improved by 10.91% compared with that of U-Net.

1.MMUU-Net

2.The comparison results of different segmentation network

(a) Original image, (b) Ground truth, (c) U-Net, (d) D-LinkNet, (e) RCF, (f) MMUU-Net (The first stage), (g) MMUU-Net (The second stage)

3.Farmland segmentation and area calculation using satellite imagery

4.On the basis of farmland segmentation, we use clustering algorithm for crop classification