Duanrui Wang, Fan Ge, Hongjian Lv, Xingyue Zhu, Dianting Zeng, Meng Wu, Jiyang Yu, Weijian Cui, Qiwei Qin, Shaowen Wang4, Chi Wu, Yi Shi
E-mail addresses: qi.wu@gmlab.ac.cn(C. Wu), shy_xflx@163.com(Y. Shi)
It has now been submitted to《Aquaculture》
🌟 Introduction
In the aquaculture industry, timely fish disease detection is vital to prevent outbreaks and minimize economic losses. DMYOLO is an enhanced model based on YOLOv10, specifically optimized for real-time fish disease detection in industrialized aquaculture environments.
📊Dataset
A custom dataset named the Belangeri Croaker Dataset was constructed by collecting images of both infected and healthy belangeri croaker. The images were captured from the Penghu semi-submersible deep-sea cage situated in proximity to Guishan Island, China. The monitoring equipment terminal was also deployed at this location. On March 19, 2024, the underwater camera was installed at a depth of 4 meters on the cage railing, and continuous image collection of belangeri croaker was conducted over a three-week period. The fish had been cultured for five months at the time of data collection.
The dataset can be downloaded from the link: https://www.kaggle.com/datasets/moonburntcat/belangeri-croaker-dataset/data
It contains folders 1-5 and labels.
🚀 Getting Started
requirements.txt: the enviroment requirements
get_FPS.py: get FPS of the model
heatmap.py: get heapmap of the model,you can choose whatever layers of the model
train_v10.py: set training parameters here and start training
test.py: test the model
The improved YAML file we have tried is located at: DMYOLO\ultralytics\cfg\models\v10
- Requirements
<git clone https://github.com/moonburntcat/DMYOLO.git
cd DMYOLO
pip install -r requirements.txt>```- Training
python train.py --model dmyolo.yaml --data data/fish.yaml --epochs 100 --batch 16- Inference / Demo
python detect.py --weights weights/best.pt --source path/to/your/video.mp4📌Contributions & Contact Us
Email us directly at [15652582873@163.com].
If you find this work useful for your research, please cite our paper.