Plastic Detection AI

Overview

This project focuses on developing an AI model capable of detecting plastic items in real-time using advanced computer vision techniques. The core model is built on YOLOv5 (You Only Look Once), a state-of-the-art object detection framework. By automating plastic detection, the project aims to address environmental challenges such as waste management and recycling inefficiencies.

The initial experiment (Exp1) uses a custom dataset of plastic items, with images and annotations provided in the data/train folder. This dataset enables the AI model to learn and effectively identify plastic items under varying conditions.

How It Works

1. Dataset:

   • The dataset is located in the data/train folder.
   • It consists of labeled images of plastic items captured in diverse environments, ensuring robust detection across real-world scenarios.

2. Model Architecture:

   • YOLOv5 is employed for its high speed and accuracy.
   • It is ideal for detecting plastic items in real-time applications like recycling plants or waste sorting systems.

3. Training:

   • The model is trained on labeled images, which include bounding boxes around plastic items.
   • Through this training, the AI learns to recognize plastic items in complex scenes.

4. Inference:

   • The trained model can detect plastic items in live video streams or static images.
   • It outputs bounding boxes, class names, and confidence scores, enabling real-time applications.

Installation and Running the Detection

To set up and run the project, follow these steps:

Step 1: Clone the repository

bash git clone <repository-url>

Step 2: Install the dependencies

bash pip install -r requirements.txt

Step 3: Download YOLOv5 weights

The trained weights (best.pt) are located in the 'runs/train' folder or can be generated by training the model.

Step 4: Train the model

bash python train.py --data data.yaml --cfg models/yolov5s.yaml --weights yolov5s.pt --epochs 100

Step 5: Run inference

python detect.py --source <video_or_image_path> --weights runs/train/best.pt

Results

In the initial experiment (Exp1), the model achieved:

Accuracy: 94% Precision: 89%

These results demonstrate that the model can reliably detect plastic items, even in challenging conditions.

Future Work

Expand the dataset to include a wider variety of plastic types and environments. Integrate the AI into recycling lines for real-time sorting and waste management. Develop an edge AI system for deployment in low-resource environments.

License

This project is licensed under the MIT License. See the LICENSE file for details.