Plastic-Net Trash Detection (an IBM Space Tech Team project)

PlasticNet is an "IBM Tech for Good" open-source project developed by the IBM Space Tech team to build a repository of AI object detection models to classify types/brands of plastics, trash on beaches, trash in the ocean, and more.

We can scale this effort with the global community of developers participating and contributing towards this noble effort, with long term goals to help with ocean cleanup and positively impact climate change.

For more information on how to get started, check out the PlasticNet Wiki!

Goals

The Goals for our project are listed below as the following:

• Real-time detection of different types of trash (plastic in particular) in the ocean utilizing transfer learning on different machine learning object detection architectures
• In the future, we would also like to be able to improve our model to be able to recognize logos/brands on trash, in order to detect and identify which company different types of ocean/beach trash come from
• To build a fully functional PlasticNet machine learning pipeline that can be easily used to train and test object detection models based from architectures such as YOLOv4, Faster-RCNN, SSD-Resnet, Efficient-DET, Tensorflow, etc. (all accessible inside a command line client)
• To provide a set of pretrained PlasticNet models that can be utilized for future development and improvement via transfer learning
• Implement our models to work on real-time satellite and camera footage

Basic Project Structure and Technologies Used (feat. diagram)

The PlasticNet command line program combines YOLOv4 and Tensorflow Object Detection API technologies into a single, easily usable machine learning pipeline CLI. Collaborators can use the PlasticNet CLI to prepare models for training (via transfer learning from the provided pre-trained PlasticNet models), train custom detection models built upon pre-trained PlasticNet models, export the trained models, and finally test the trained models. The CLI was created so these steps can all be done with a few simple commands.

Initially trained via transfer learning from pre-trained YOLO weights, and pre-trained Tensorflow models from the Tensorflow Detection Model Zoo, our official PlasticNet Model Zoo can be used by collaborators for the further improvement/development of new PlasticNet object detection models. For labeling images, we utilized IBM's Cloud Annotations.

Demo of Object Detection

YOLOv4 9 class v3 PlasticNet Demo (best YOLO model, with face masks and fishing nets included):

Tensorflow EfficientDET D1 9 Class Demo:

Get Started

To get started with PlasticNet, you first must clone the repository using the following command:

git clone https://github.com/IBM/PlasticNet.git

For more detailed instructions on how to get started, check out the PlasticNet Wiki!

Once the repository is cloned, you can run the following command in the python evironment of your choice. (NOTE: It is recommended that this is done in a new python environment, to avoid any issues between package dependencies)

The setup script currently only supports MacOS, but Windows and Linux support will be added soon.

cd PlasticNet && python setup.py

Once the setup script has finished running, you should restart your terminal session and run the following command:

PlasticNet

This will open the PlasticNet terminal, so you can easily download our models from our model zoo, test the models on videos, webcam, and images, or train on top of an existing model.

A list of all commands can be found by typing help, and more detailed instructions about aguments for any command can be found with help [command name].

To exit the command line, type quit.

Using YOLO Models

If you intend to train YOLO Models, you, may have to make some changes to the Makefile depending on your system. The Makefile is located in ./darknet, and you will want to update these parameters to whatver your system has, 1 meaning enabled.

GPU=1
CUDNN=0
CUDNN_HALF=0
OPENCV=1
AVX=0
OPENMP=0
LIBSO=0

It is highly recommended you install CUDNN & OpenCV for use with Darknet, as it will expedite the training process.

You can additionally update the yolo-obj.cfg file (located in /darknet/cfg/) with any parameters you choose, specifically the number of iterations, classes, and filters. Here's a guide for setting these values.

After you have updated this makefile and/or the configuration file, run make clean and make to be able to start darknet training with the PlasticNet CLI.

Test Results

See our spreadsheet documentating our test results from different trained models.

Resources

• Forked YOLOv4 PlasticNet Darknet repository
• Labeling Images with IBM Cloud Annotations
• Darknet YOLO
• Restoring Integrity to the Oceans (RIO)
• Pacific Whale Foundation
• Tensorflow Object Detection API
• Yolo with Tensorflow

Ocean Plastic Statistics

• More than 1 million seabirds and 100,000 marine animals die from plastic pollution every year.
• 100% of baby sea turtles have plastic in their stomachs.
• There is now 5.25 trillion macro and micro pieces of plastic in our ocean & 46,000 pieces in every square mile of ocean, weighing up to 269,000 tonnes.
• Every day around 8 million pieces of plastic makes their way into our oceans.
• The Great Pacific Garbage Patch is around 1.6 million square kilometers – bigger than Texas.
• The world produces 381 million tonnes in plastic waste yearly – this is set to double by 2034.
• 50% of this is single-use plastic & only 9% has ever been recycled.
• Over 2 million tonnes of plastic packaging are used in the UK each year.
• 88% of the sea's surface is polluted by plastic waste.
• Between 8 to 14 million tonnes enters our ocean every year.
• Britain contributes an estimated 1.7 million tonnes of plastic annually.
• The US contributes 38 million tonnes of plastic every year.
• Plastic packaging is the biggest culprit, resulting in 80 million tonnes of waste yearly from the US alone.
• On UK beaches there are 5000 pieces of plastic & 150 plastic bottles for each mile.
• More than 1 million plastic bags end up in the trash every minute.
• The world uses over 500 billion plastic bags a year – that’s 150 for each person on Earth.
• 8.3 billion plastic straws pollute the world’s beaches, but only 1% of straws end up as waste in the ocean.
• By 2020 the number of plastics in the sea will be higher than the number of fish.
• 1 in 3 fish caught for human consumption contains plastic.
• Plastic microbeads are estimated to be one million times more toxic than the seawater around it.
• Products containing microbeads can release 100,000 tiny beads with just one squeeze.

Source: Condor Ferries: Shocking Ocean Plastic Statistics