https://github.com/arssite/dirty-cleanflooringimageprocessingusingyolov5
Uses YOLOv5 to classify floor cleanliness into five categories based on visual cues. It includes an annotated dataset, trained model,& evaluation outputs. Code covers data preprocessing, training, & testing. A comparative analysis highlights YOLOv5's advantages over traditional methods, providing an efficient solution automated floor cleanliness.
https://github.com/arssite/dirty-cleanflooringimageprocessingusingyolov5
deep-neural-networks github google-colab jupyter-notebook labelimg matplotlib-pyplot numpy-library opencv-python pandas-python pytorch scikit-learn tensorflow yolov5
Last synced: 7 months ago
JSON representation
Uses YOLOv5 to classify floor cleanliness into five categories based on visual cues. It includes an annotated dataset, trained model,& evaluation outputs. Code covers data preprocessing, training, & testing. A comparative analysis highlights YOLOv5's advantages over traditional methods, providing an efficient solution automated floor cleanliness.
- Host: GitHub
- URL: https://github.com/arssite/dirty-cleanflooringimageprocessingusingyolov5
- Owner: arssite
- License: cc0-1.0
- Created: 2024-11-25T20:43:20.000Z (11 months ago)
- Default Branch: main
- Last Pushed: 2024-11-26T12:18:48.000Z (11 months ago)
- Last Synced: 2025-02-01T01:30:58.936Z (8 months ago)
- Topics: deep-neural-networks, github, google-colab, jupyter-notebook, labelimg, matplotlib-pyplot, numpy-library, opencv-python, pandas-python, pytorch, scikit-learn, tensorflow, yolov5
- Language: Jupyter Notebook
- Homepage:
- Size: 30.4 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Clean vs. Dirty Floor Identification Using YOLOv5
## Project Overview
This project aims to develop an algorithm that identifies and rates the cleanliness of floors using visual cues. The algorithm assigns a cleanliness score ranging from 0 (dirty) to 4 (clean). YOLOv5, a state-of-the-art object detection model, is used for this task due to its high speed, accuracy, and scalability.
---
## Why YOLO is the Best Choice?
| Feature | YOLOv5 | Other Methods |
|-----------------------------------|----------------------------------|----------------------------------|
| **Accuracy** | High | Varies |
| **Speed** | Real-time detection | Slower |
| **Scalability** | Supports large datasets | Limited for traditional CV |
| **Ease of Integration** | Pretrained weights available | Requires custom development |
YOLOv5 is chosen because:
- It offers **real-time detection**, which is essential for dynamic environments.
- **Pretrained weights** allow faster prototyping and deployment.
- It excels in tasks with **complex datasets**, handling multiple object categories efficiently.
---
## Directory Structure
```plaintext
project/
├── dataset/
│ ├── images/
│ │ ├── train/ # Training images
│ │ ├── val/ # Validation images
│ │ └── test/ # Test images
│ ├── labels/
│ │ ├── train/ # Annotations for training
│ │ ├── val/ # Annotations for validation
│ │ └── test/ # Annotations for test set
├── yolo/
│ ├── config/ # YOLO model configuration
│ ├── weights/ # YOLO pretrained weights
│ ├── output/ # Outputs (trained model, results)
│ └── logs/ # Training logs
```
---
## Comparison of Methods
| **Method** | **Accuracy** | **Speed** | **Scalability** | **Dataset Requirements** |
|---------------------------|--------------|------------|-----------------|-----------------------------------|
| Classical CV (e.g., SVM) | Moderate | High | Low | Small |
| CNN (Custom/Pretrained) | High | Moderate | High | Large |
| Semantic Segmentation | Very High | Moderate | High | Pixel-Annotated Dataset |
| Faster R-CNN/Mask R-CNN | Very High | Low | High | Moderate |
| K-Means Clustering | Moderate | High | Low | Unlabeled Data |
| Multi-Layer Perceptron | Moderate | High | Low | Feature-Based Dataset |
| **YOLO (Chosen)** | High | Real-time | High | Moderate |
---
## Example Annotated Output
Annotated images include bounding boxes and cleanliness scores (0–4). Below is an example of a processed image:
---
## Steps to Reproduce
1. **Data Collection**:
- Collect images of clean and dirty floors under varying conditions (lighting, floor types).
- Manually annotate the images with cleanliness scores (0–4).
2. **Dataset Preparation**:
- Organize images and labels as per the directory structure.
- Use tools like LabelImg for annotation.
3. **Training**:
- Train YOLOv5 using the prepared dataset and configurations.
- Command:
```bash
python train.py --img 640 --batch 16 --epochs 50 --data dataset.yaml --weights yolov5s.pt
```
4. **Validation**:
- Validate the trained model.
- Command:
```bash
python val.py --data dataset.yaml --weights runs/train/exp/weights/best.pt
```
5. **Testing**:
- Test the model on new images.
- Command:
```python
def test_image(image_path):
# Function to load, process, and annotate image
```
6. **Saving Outputs**:
- Save annotated outputs using the provided function.
---
## Metrics Evaluation
Key performance metrics:
- **Accuracy**: High (~90%+ depending on dataset quality)
- **Precision**: High, ensuring fewer false positives.
- **Recall**: Adequate, reducing missed detections.
- **F1-Score**: Balanced metric combining precision and recall.
---
## Future Improvements
- Incorporate **data augmentation** for robustness.
- Experiment with **advanced architectures** like YOLOv8 or custom CNNs.
- Add real-time deployment capabilities using NVIDIA TensorRT.
---
## References (Relevant research papers or resources)
https://www.nature.com/articles/s41598-023-38538-3.pdf
file:///C:/Users/hp/Downloads/ADeepLearningApproachforClassificationofCleanlinessinRestrooms.pdf
https://www.researchgate.net/publication/356691001_A_Deep_Learning-Based_Dirt_Detection_Computer_Vision_System_for_Floor-Cleaning_Robots_with_Improved_Data_Collection