MIMO-SAR Based Soil Wetness Detection

Imaging Result

Video

MIMO-SAR Imaging of Hallow Square Iron Piece:

https://youtube.com/shorts/pGSzS-oHYfA?feature=share

MIMO-SAR Imaging of Soil:

https://youtube.com/shorts/OX6Q5ki3xco?feature=share

Hardware:

• Driver: DM542S
• Motor : 42 BYG H4812 Motor
• Microcontroller: Ardiuno Mega
• mmWave Radar: AWR1843 Boost
• 400 $\times$ 400 2D Scanner
• 4mm pitch of thread and 9mm diameter

Software:

• mmWave Studio
• Ardiuno IDE
• Matlab
• Python

MIMO-SAR-mmWave-Imaging-Toolbox

In this repository, we leverage and modify a comprehensive open-source MIMO-SAR imaging toolbox, which is a MATLAB based software package.

Citation: Yanik, Muhammet Emin, et al. “Development and Demonstration of MIMO-SAR MmWave Imaging Testbeds.” IEEE Access, vol. 8, Institute of Electrical and Electronics Engineers (IEEE), 2020, pp. 126019–38, doi:10.1109/access.2020.3007877.

Our code consists of three main modules: (1) data capture, (2) MIMO array calibration, and (3) image reconstruction. The framework of each module is detailed in our journal paper.

How to Use

Data Collection or Data Access

You can build up one platform leveraging all the hardware components mentioned above and collect your own MIMO-SAR data, which may be quite challenging and exhausting.

Therefore, we provide you with the following data to make you have a general idea about what the imaging is like.

You can access these data shared by Baidu Disk ：MIMO-SAR Data through the following link.

• Link: https://pan.baidu.com/s/1aAp8LxT2ngH5ruKbu5GJuw?pwd=3ww1

• Password: 3ww1

Modify all Necessary directories and files

Put all the data you collect or download into the file path ' .\Data\valid'

%--------------------------------------------------------------------------
% Path for the adcBinData
experiment_folderName = '.\Data';
adcBinData_subFolderName = 'valid';

% Path for the Calibration Data 
calData_fileName = '.\Algorithms\calibration\calData\calData_3Tx_4Rx.mat';
delayOffset_fileName = '.\Algorithms\calibration\calData\delayOffset_3Tx_4Rx.mat';

%Here we can also add path for background calibration 

Mean Target z-Distance and Other Modified Parameters

zTarget_mm = 250; % update the distance as needed.

sensorParams.Num_Frames = 400;

sarParams.Num_verticalScan = 50 %Update these two parameters as well

Choose the data you read

Open the file .\Algorithms\dataRead\fcn_read_ADC_bin_SAR.m

Change the adc_file_name to the name of your data

function [rawData_Rxchain]= fcn_read_ADC_bin_SAR(sensorParams,sarParams,adcBinData_folderName)

Num_RX_channels = 4; % Fixed for this version

Samples_per_Chirp = sensorParams.Samples_per_Chirp;

Chirps_per_Frame = sensorParams.Chirps_per_Frame;

Num_TX = sensorParams.Num_TX;

Num_Frames = sensorParams.Num_Frames;

Num_measurements = sarParams.Num_horizontalScan * sarParams.Num_verticalScan;

% adc_file_name = [adcBinData_folderName '\ADC Temp.bin'];

% adc_file_name = [adcBinData_folderName '\sar_data_con_3_Raw_0.bin'];

adc_file_name = [adcBinData_folderName '\dry.bin'];

% [ErrStatus,rawData] = Parse_Datafile_bin_SAR(adc_file_name,2*Num_RX_channels,Samples_per_Chirp,Chirps_per_Frame*Num_TX,Num_Frames,Num_measurements);

% if ErrStatus~=0

%     disp('Error in parsing data file for Rxchain');

%     return

% end

rawData = read1843Data(adc_file_name,256,4);

% //I+Q:2-> complex

% 

% rawData = reshape(rawData,2*Num_RX_channels,[]);

% 

% rawData1 = rawData(1,:) + 1i*rawData(3,:);

% rawData2 = rawData(2,:) + 1i*rawData(4,:);

% rawData3 = rawData(5,:) + 1i*rawData(7,:);

% rawData4 = rawData(6,:) + 1i*rawData(8,:);

% rawData_Rxchain = [rawData1;rawData2;rawData3;rawData4];

% rawData_Rxchain = rawData([1,3,5,7],:) + 1i*rawData([2,4,6,8],:);

% reshape the rawData_Rxchain

rawData_Rxchain = reshape(rawData,Num_RX_channels,Samples_per_Chirp,Num_TX,Chirps_per_Frame,Num_Frames,Num_measurements);

% reshape the rawData_Rxchain as Num_RX_channels x Num_TX x Samples_per_Chirp x Chirps_per_Frame x Num_Frames x Num_measurements

rawData_Rxchain = permute(rawData_Rxchain,[1,3,2,4,5,6]);

Run the workflow

After modify the directories, files and parameters, running the workflow is what you need !

Training and testing

You are supposed to start training only if you arrange your data well. A brief structure is as follows.

data/
├── train/
│   ├── class1/
│   │   ├── image1.jpg
│   │   ├── image2.jpg
│   │   └── ...
│   ├── class2/
│   └── ...
└── val/
    ├── class1/
    │   ├── image1.jpg
    │   ├── image2.jpg
    │   └── ...
    ├── class2/
    └── ...

Here, class1, class2 and etc. are regarded as labels of each class.

Then, run the following line in your command to install necessary dependencies if you have not.

pip install torch torchvision tqdm

And run the script train.py to start. GPU is supported and highly recommended for efficient training.

python train.py

Finally, after getting your model [name].pth, just modify the dict CIFAR10_CLASSES to match the class No. and the label, and set num_classes correctly. Run the last line after you name the target file test_img/tst_image.png and you will get your result.

python predict.py