Overview / Usage

How we can differentiate garbage types whether "bottle", "pet bottle", "burnable" and so on? we sometimes hard to differentiate it. My experience when visited Japan, there are several waste bin types like i mentioned before. which one is correct when i have "banana peel" ? . In my country also has several types of waste bin types like "organic" and "not organic" , which one is correct when i have "bottle"? we don't know.
So this project is created to differentiate garbage type based on images.

how it works? the user just upload image of garbage to system and the system will response the kind of garbages correctly! the garbage image will be forwarded to inference machine that i prepared before. the model in inference machine is Convolutional Neural Network. this system accuracy in predicting garbage type reach 90%.

Step-by-step of implementation:

1. Download dataset from Image-net Stanford (imagenet.stanford.edu). This dataset contains 1000 class of images. we select only image that appropriate for our system like: Banana, Bottle, Pet Bottle, and so on. we additionally augment other image from other source to enrich dataset

2. Train the model on dataset based on Convolutional Neural Network (see my previous project: https://devmesh.intel.com/projects/landuse-classification-convolutional-neural-network)

3. After obtaining a model, we try to inference a new image to model. let say image captured by camera will be forwarded to this machine.

4. create Front-end like Android or web apps that perform capturing images and forwarded to model. this sometimes use web service to connect front-end with model. in this case i use Flask micro-framework in Python environment as web apps and web service.