Ship detection in SAR image based on yolov5

stay yolov5 Environment building After completion , You can train the model by yourself , This article focuses on the use of yolov5 Realization SAR Detection of ships in images .

1. Data set production

1.1 Download datasets

SSDD The data set can be in Baidu SkyDrive , Data set comparison , Don't wait too long
link ：https://pan.baidu.com/s/1aXeDJ1-bd3-Wftgk4_7glg
Extraction code ：2022

1.2 Dataset format

What we have on hand SSDD The data set format is as follows ：

- SSDD
	- Annotations
		- 000001.xml
		- 000002.xml
		- ...
	- JPEGImages
		- 000001.JPG
		- 000002.JPG
		- ...

among ,Annotations Labels for all pictures , every last xml The tag file contains the target type , The location of the target （ The center position, width and height of the frame ）

---

YOLOV5 Datasets in this format are not supported , We can modify the data set to COCO Format ：

- ships
	- images
		- train (JPG images)
			- 000341.JPG
			- 000024.JPG
			- ...
		- test (JPG images)
		- val (JPG images) 
	- labels
		- train (txt files)
			- 000341.txt
			- 000024.txt
			- ...
		- test (txt files)
		- val (txt files)
		- ...

You can see , Data set containing /images and /labels,images in It contains three folders to store three types of data sets , Corresponding labels Three types of labels are also stored in , It is worth noting that ,labels Not in yes xml file , It is txt file , every last txt The contents of the document are as follows ：

 Format ：<object-class> <x> <y> <width> <height>
0  0.41416  0.61538  0.31607  0.11242

---

1.3 format conversion

It's more difficult to deal with , I wrote a function , Can be placed with the original SSDD Folder in the same directory , Automatically generate corresponding after running COCO Data set in format ships_dataset, Code download address ：

link ：https://pan.baidu.com/s/1GMODNViP0WaquqJUyme_YA
Extraction code ：2022

The effect after operation is as follows ：

You can generate ships_dataset Throw it yolov5 In the table of contents , To be used later .

---

1.4 establish yaml file

stay yolov5 You can find it in the source code directory coco Data sets yaml file , It is used to describe the location of the data set and the number of sample types included ：

The contents are as follows ：

# YOLOv5  by Ultralytics, GPL-3.0 license
# COCO128 dataset https://www.kaggle.com/ultralytics/coco128 (first 128 images from COCO train2017) by Ultralytics
# Example usage: python train.py --data coco128.yaml
# parent
# ├── yolov5
# └── datasets
# └── coco128 ← downloads here (7 MB)


# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
path: ../datasets/coco128  # dataset root dir
train: images/train2017  # train images (relative to 'path') 128 images
val: images/train2017  # val images (relative to 'path') 128 images
test:  # test images (optional)

# Classes
nc: 80  # number of classes
names: ['person', 'bicycle', 'car', ...]  # class names

among ,path It points out the directory where the data set is stored ,train Point out the training set picture The position of ,val Indicates the validation set picture The position of ,test It's not necessary .

Corresponding , We created ships.yaml, The contents are as follows ：

# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
path: ./ships_dataset  # dataset root dir
train: images/train  # train images (relative to 'path') 128 images
val: images/val  # val images (relative to 'path') 128 images
test: images/test  # test images (optional)

# Classes
nc: 1  # number of classes
names: ['ship']  # class names

---

2. debugging

Use the following command to train the model ：

python train.py --img 320 --batch-size 8 --epochs 10 --data data/ships.yaml --cfg models/yolov5s.yaml --weights yolov5s.pt --device 0

For parameter descriptions in the command, please refer to Official explanation and This blog , Notice here that we have just created yaml file （ These parameters can be found in train.py Set the default value in , You can also specify... In the command ）

First training , Need to download some small files , The code automatically passes Github Download from the official website , It may be due to inaccessibility Github Report errors , Change code directly ：

take yolov5/utils/downloads.py All in github.com Change to mirror URL hub.xn--p8jhe.tw

response = requests.get(f'https://hub.xn--p8jhe.tw/repos/

In this way, it can be executed normally , run 10 individual epoch Some ships can be detected , But the accuracy is not high ,100 epoches Can basically achieve 94% About the detection rate .

3. Optimize

Because it's still Xiaobai , There is no experience in adjusting parameters , At present, we first improve the accuracy through image enhancement , Reference resources Blog

stay yolov5\data\hyps You can see that there are many files related to super parameter settings , Create your own hyp.ships.yaml, Copy first hyp.scratch-low.yaml The whole content of , Value to modify the image enhancement part ：

hsv_h: 0.015  # image HSV-Hue augmentation (fraction)
hsv_s: 0.5  # image HSV-Saturation augmentation (fraction)
hsv_v: 0.3  # image HSV-Value augmentation (fraction)
degrees: 0  # image rotation (+/- deg)
translate: 0.1  # image translation (+/- fraction)
scale: 0.75  # image scale (+/- gain)
shear: 0.0  # image shear (+/- deg)
perspective: 0.003  # image perspective (+/- fraction), range 0-0.001
flipud: 0.4  # image flip up-down (probability)
fliplr: 0.5  # image flip left-right (probability)
mosaic: 1.0  # image mosaic (probability)
mixup: 0.0  # image mixup (probability)
copy_paste: 0.0  # segment copy-paste (probability)

After using image enhancement, you need to train.py Set the parameter value to

parser.add_argument('--hyp', type=str, default=ROOT / 'data/hyps/hyp.ships.yaml', help='hyperparameters path')

3. Training and results

After using image enhancement , Retraining 150 individual epoches（ wait for 3 Hours ）：

150 epochs completed in 3.510 hours.
Optimizer stripped from runs\train\exp40\weights\last.pt, 14.3MB
Optimizer stripped from runs\train\exp40\weights\best.pt, 14.3MB

Validating runs\train\exp40\weights\best.pt...
Fusing layers...
YOLOv5s summary: 213 layers, 7012822 parameters, 0 gradients, 15.8 GFLOPs
               Class     Images     Labels          P          R     [email protected] [email protected]:.95: 100%|██████████| 15/15 [00:04<0
                 all        232        578       0.95      0.865      0.947      0.526
Results saved to runs\train\exp40

Can be in yolov5/runs/train/exp40 View results ：

You can also take the trained weight/best.pt File predictions （ Next, predict the pictures in all test sets ）：

python detect.py --iou-thres 0.3 --weights ./runs/train/exp40/weights/best.pt --source ./ships_dataset/images/test --save-txt --save-crop

Can be in yolov5/runs/detect View results